A method, device and electronic device for processing aerial scene images

Abstract

The present application provides a method for processing an aerial scene image, comprising: obtaining a target real aerial scene image, and obtaining a simulated aerial scene image to be processed, wherein the simulated aerial scene image to be processed is a simulated image corresponding to the target real aerial scene image; inputting the simulated aerial scene image to be processed and the target real aerial scene image into a target generative adversarial network model, and obtaining a target aerial scene simulated image whose image feature data matches the image feature data of the target real aerial scene image. The method for processing an aerial scene image provided in the present application can obtain a target aerial scene simulated image whose image feature data matches the image feature data of the target real aerial scene image, thereby improving the authenticity of the aerial scene simulated image.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more particularly to a method for processing an aerial scene image. The present application also relates to an aerial scene image processing device, electronic equipment, and storage medium. Background Art

[0002] Deep neural network models have been widely used in image classification, detection and related fields. In the training process of deep neural network models, a large number of real images are often required as training sample images. For example, when training a deep neural network model for processing scene images corresponding to actual scenes, a large number of real scene images need to be obtained as training sample images.

[0003] In general, there are relatively few publicly available and free real scene images that can be used directly, especially aviation-related aerial scene images. Therefore, in order to obtain a large number of aerial scene images as training sample images for training deep neural network models for processing aerial scene images, it is necessary to synthesize aerial scene images through image synthesis. When synthesizing aerial scene images through existing image synthesis methods, there is often a problem of low authenticity of aerial scene images. Summary of the invention

[0004] The present application provides a method, device, electronic device and storage medium for processing an aviation scene image, so as to improve the realism of an aviation scene simulation image.

[0005] The present application provides a method for processing an aerial scene image, comprising:

[0006] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0007] The to-be-processed aviation scene simulation image and the target real aviation scene image are input into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0008] Optionally, the step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises:

[0009] Obtaining a sample aerial scene simulation image and obtaining a sample real aerial scene image;

[0010] The to-be-trained generative adversarial network model is trained according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model.

[0011] Optionally, the performing model training on the to-be-trained generative adversarial network model according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model includes:

[0012] Inputting a first sample aviation scene simulation image and a first sample real aviation scene image into the to-be-trained generative adversarial network model to obtain image feature data of the first sample real aviation scene image;

[0013] Migrating the image feature data of the first sample real aviation scene image to the first sample aviation scene simulation image to obtain a first target aviation scene simulation image;

[0014] Determining whether the image feature data of the first target aviation scene simulation image matches the image feature data of the first sample real aviation scene image;

[0015] If so, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

[0016] Optionally, also include:

[0017] If the image feature data of the first target aviation scene simulation image does not match the image feature data of the first sample real aviation scene image, inputting the second sample aviation scene simulation image and the second sample real aviation scene image into the to-be-trained generative adversarial network model to obtain the image feature data of the second sample real aviation scene image;

[0018] Migrating the image feature data of the second sample real aviation scene image to the second sample aviation scene simulation image to obtain a second target aviation scene simulation image;

[0019] Determining whether the image feature data of the second target aviation scene simulation image matches the image feature data of the second sample real aviation scene image;

[0020] If the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

[0021] Optionally, it also includes: if the image feature data of the second target aerial scene simulation image does not match the image feature data of the second sample real aerial scene image, then inputting the sample aerial scene simulation image and the sample real aerial scene image into the generative adversarial network model to be trained in sequence until the target generative adversarial network model is obtained.

[0022] Optionally, the step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises:

[0023] Inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real aerial scene image;

[0024] The high-dimensional image feature data of the target real aviation scene image is transferred into the aviation scene simulation image to be processed to obtain the target aviation scene simulation image.

[0025] Optionally, the step of transferring the high-dimensional image feature data of the target real aerial scene image into the to-be-processed aerial scene simulation image to obtain the target aerial scene simulation image includes:

[0026] Obtaining initial high-dimensional image feature data of the to-be-processed aerial scene simulation image;

[0027] According to the high-dimensional image feature data of the target real aviation scene image, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image is adjusted to obtain the target aviation scene simulation image.

[0028] Optionally, adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image includes:

[0029] According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain a candidate aviation scene simulation image;

[0030] Performing clustering comparison on the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed;

[0031] If the similarity reaches or exceeds a similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

[0032] Optionally, also include:

[0033] If the similarity does not reach the similarity threshold, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image continues to be adjusted to obtain candidate aviation scene simulation images until the target aviation scene simulation image is obtained.

[0034] Optionally, the high-dimensional image feature data of the target real aerial scene image includes at least one or more of image texture feature data, image clarity feature data, image color saturation feature data, image contrast feature data and image brightness data.

[0035] Optionally, obtaining a target real aerial scene image and obtaining a simulated aerial scene image to be processed includes:

[0036] Obtaining a target object in the target real aerial scene image;

[0037] Constructing a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object;

[0038] Perform imaging simulation on the three-dimensional scene model corresponding to the aviation scene simulation image to be processed to obtain the aviation scene simulation image to be processed.

[0039] Optionally, constructing a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object includes:

[0040] According to the target real aerial scene image, obtaining a target two-dimensional image of a three-dimensional object corresponding to the target object through a designated image acquisition device;

[0041] Establishing a target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device;

[0042] Extracting image feature data of the target two-dimensional image;

[0043] Determining a corresponding relationship of the target object in the three-dimensional space coordinate system according to the image feature data of the target two-dimensional image;

[0044] The three-dimensional scene model is constructed in the target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system.

[0045] Optionally, constructing the three-dimensional scene model in the target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system includes:

[0046] Determining, according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system, the sparse point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system;

[0047] Performing point cloud expansion on the sparse point cloud data to obtain dense point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system;

[0048] The three-dimensional scene model is obtained and constructed according to the dense point cloud data.

[0049] Optionally, performing imaging simulation on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed to obtain the aerial scene simulation image to be processed includes: performing imaging simulation on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed according to parameter data of a designated image acquisition device and preset lighting parameter data to obtain the aerial scene simulation image to be processed.

[0050] Optionally, the method further includes: obtaining parameter data of the designated image acquisition device.

[0051] Optionally, also include:

[0052] Obtaining annotation data of objects in the aerial scene simulation image to be processed;

[0053] Obtaining a matching relationship between a target object in the target aerial scene simulation image and the object;

[0054] The labeled data of the target object is obtained according to the labeled data of the object and the matching relationship.

[0055] Optionally, the simulated aerial scene image to be processed is a simulated image corresponding to the target real aerial scene image, including: the simulated aerial scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aerial scene image.

[0056] In another aspect, the present application provides a device for processing an aerial scene image, comprising:

[0057] An image acquisition unit, used to acquire a target real aerial scene image and to acquire a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0058] The image processing unit is used to input the to-be-processed aviation scene simulation image and the target real aviation scene image into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0059] In another aspect, the present application provides an electronic device, including:

[0060] Processor; and

[0061] The memory is used to store a program of a method for processing an aerial scene image. After the device is powered on and the program of the method for processing an aerial scene image is run by the processor, the following steps are performed:

[0062] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0063] The to-be-processed aviation scene simulation image and the target real aviation scene image are input into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0064] In another aspect of the present application, a storage medium is provided, which stores a program of a method for processing an aerial scene image, and the program is executed by a processor to perform the following steps:

[0065] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0066] The to-be-processed aviation scene simulation image and the target real aviation scene image are input into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0067] In another aspect, the present application provides a method for processing an aerial scene image, comprising:

[0068] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0069] Obtaining annotation data of objects in the aerial scene simulation image to be processed;

[0070] Obtaining a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image;

[0071] The labeling data of the target object in the target aerial scene simulation image is obtained according to the labeling data of the object.

[0072] Optionally, obtaining the labeling data of the target object in the target aerial scene simulation image according to the labeling data of the object includes:

[0073] Obtaining a matching relationship between the target object and the object;

[0074] The labeled data of the target object is obtained according to the labeled data of the object and the matching relationship.

[0075] Optionally, the annotation data of the object includes at least one or more of the position data of the object in the aerial scene simulation image to be processed, the name data of the object and the contour data of the object.

[0076] Optionally, obtaining a target aviation scene simulation image based on the aviation scene simulation image to be processed and the target real aviation scene image comprises: inputting the aviation scene simulation image to be processed and the target real aviation scene image into a target generative adversarial network model to obtain the target aviation scene simulation image.

[0077] Optionally, the step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises:

[0078] Obtaining a sample aerial scene simulation image and obtaining a sample real aerial scene image;

[0079] The to-be-trained generative adversarial network model is trained according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model.

[0080] Optionally, the performing model training on the to-be-trained generative adversarial network model according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model includes:

[0081] Inputting a first sample aviation scene simulation image and a first sample real aviation scene image into the to-be-trained generative adversarial network model to obtain image feature data of the first sample real aviation scene image;

[0082] Migrating the image feature data of the first sample real aviation scene image to the first sample aviation scene simulation image to obtain a first target aviation scene simulation image;

[0083] Determining whether the image feature data of the first target aviation scene simulation image matches the image feature data of the first sample real aviation scene image;

[0084] If so, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

[0085] Optionally, also include:

[0086] If the image feature data of the first target aviation scene simulation image does not match the image feature data of the first sample real aviation scene image, inputting the second sample aviation scene simulation image and the second sample real aviation scene image into the to-be-trained generative adversarial network model to obtain the image feature data of the second sample real aviation scene image;

[0087] Migrating the image feature data of the second sample real aviation scene image to the second sample aviation scene simulation image to obtain a second target aviation scene simulation image;

[0088] Determining whether the image feature data of the second target aviation scene simulation image matches the image feature data of the second sample real aviation scene image;

[0089] If the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

[0090] Optionally, it also includes: if the image feature data of the second target aerial scene simulation image does not match the image feature data of the second sample real aerial scene image, then inputting the sample aerial scene simulation image and the sample real aerial scene image into the generative adversarial network model to be trained in sequence until the target generative adversarial network model is obtained.

[0091] Optionally, the step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises:

[0092] Inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real aerial scene image;

[0093] The high-dimensional image feature data of the target real aviation scene image is transferred into the aviation scene simulation image to be processed to obtain the target aviation scene simulation image.

[0094] Optionally, the step of transferring the high-dimensional image feature data of the target real aerial scene image into the to-be-processed aerial scene simulation image to obtain the target aerial scene simulation image includes:

[0095] Obtaining initial high-dimensional image feature data of the to-be-processed aerial scene simulation image;

[0096] According to the high-dimensional image feature data of the target real aviation scene image, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image is adjusted to obtain the target aviation scene simulation image.

[0097] Optionally, adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image includes:

[0098] According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain a candidate aviation scene simulation image;

[0099] Performing clustering comparison on the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed;

[0100] If the similarity reaches or exceeds a similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

[0101] Optionally, the obtaining of the target real aerial scene image and the obtaining of the to-be-processed aerial scene simulation image comprises: obtaining a target object in the target real aerial scene image;

[0102] Constructing a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object;

[0103] Performing imaging simulation on the three-dimensional scene model corresponding to the aviation scene simulation image to be processed to obtain the aviation scene simulation image to be processed;

[0104] The obtaining of the annotation data of the object in the to-be-processed aviation scene simulation image includes: obtaining the annotation data of the object according to a three-dimensional scene model corresponding to the to-be-processed aviation scene simulation image.

[0105] Optionally, the simulated aerial scene image to be processed is a simulated image corresponding to the target real aerial scene image, including: the simulated aerial scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aerial scene image.

[0106] In another aspect, the present application provides a device for processing an aerial scene image, comprising:

[0107] An image acquisition unit, used to acquire a target real aerial scene image and to acquire a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0108] A labeling data obtaining unit, used for obtaining labeling data of objects in the aerial scene simulation image to be processed;

[0109] A target image acquisition unit, configured to acquire a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image;

[0110] The annotation data obtaining unit is used to obtain the annotation data of the target object in the target aerial scene simulation image according to the annotation data of the object.

[0111] In another aspect, the present application provides an electronic device, including:

[0112] Processor; and

[0113] The memory is used to store a program of a method for processing an aerial scene image. After the device is powered on and the program of the method for processing an aerial scene image is run by the processor, the following steps are performed:

[0114] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0115] Obtaining annotation data of objects in the aerial scene simulation image to be processed;

[0116] Obtaining a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image;

[0117] The labeling data of the target object in the target aerial scene simulation image is obtained according to the labeling data of the object.

[0118] In another aspect of the present application, a storage medium is provided, which stores a program of a method for processing an aerial scene image, and the program is executed by a processor to perform the following steps:

[0119] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0120] Obtaining annotation data of objects in the aerial scene simulation image to be processed;

[0121] Obtaining a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image;

[0122] The labeling data of the target object in the target aerial scene simulation image is obtained according to the labeling data of the object.

[0123] In another aspect, the present application provides an image processing method, comprising:

[0124] Obtaining a target real image and obtaining a simulated image to be processed, wherein the simulated image to be processed is a simulated image corresponding to the target real image;

[0125] The simulated image to be processed and the target real image are input into a target generative adversarial network model to obtain a target simulated image whose image feature data matches the image feature data of the target real image.

[0126] In another aspect, the present application provides an image processing method, comprising:

[0127] Obtaining a target real image and obtaining a simulated image to be processed, wherein the simulated image to be processed is a simulated image corresponding to the target real image;

[0128] Obtaining annotation data of objects in the simulated image to be processed;

[0129] Obtaining a target simulated image according to the simulated image to be processed and the target real image;

[0130] The labeling data of the target object in the target simulation image is obtained according to the labeling data of the object.

[0131] Compared with the prior art, this application has the following advantages:

[0132] The present application provides a method for processing aerial scene images. After obtaining a target real aerial scene image and a to-be-processed aerial scene simulation image generated by image simulation and corresponding to the target real aerial scene image, the to-be-processed aerial scene simulation image and the target real aerial scene image are input into a trained target generative adversarial network model to further obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image, thereby improving the authenticity of the aerial scene simulation image.

[0133] Another method for processing aerial scene images provided by the present application further obtains annotation data of objects in the aerial scene simulation image to be processed after obtaining the aerial scene simulation image to be processed corresponding to the target real aerial scene image. Then, after obtaining the target aerial scene simulation image based on the aerial scene simulation image to be processed and the target real aerial scene image, the annotation data of the target object in the target aerial scene simulation image can be further obtained based on the annotation data of the object, thereby ensuring that the obtained target aerial scene simulation image can carry the annotation data of the target object in the target aerial scene simulation image, thereby realizing automatic data annotation of the target object in the target aerial scene simulation image and simplifying the process of data annotation of aerial scene images. BRIEF DESCRIPTION OF THE DRAWINGS

[0134] Figure 1 It is the first schematic diagram of the application scenario of the aerial scene image processing method provided in this application.

[0135] Figure 2 This is a second schematic diagram of an application scenario of the method for processing aerial scene images provided in this application.

[0136] Figure 3 It is a flowchart of the method for processing aerial scene images provided in the first embodiment of the present application.

[0137] Figure 4 It is a flowchart of a method for generating a target aerial scene simulation image provided in the first embodiment of the present application.

[0138] Figure 5 It is a schematic diagram of an aerial scene image processing device provided in the second embodiment of the present application.

[0139] Figure 6 It is a schematic diagram of an electronic device provided in an embodiment of the present application.

[0140] Figure 7 It is a flowchart of a method for processing an aerial scene image provided in the fifth embodiment of the present application.

[0141] Figure 8 It is a schematic diagram of an aerial scene image processing device provided in the sixth embodiment of the present application. DETAILED DESCRIPTION

[0142] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0143] In order to more clearly demonstrate the method for processing aerial scene images provided by the present application, the application scenario of the method for processing aerial scene images provided by the present application is first introduced.

[0144] The aerial scene image processing method provided in this application is generally used to generate a scene with an aerial scene simulation image carrying annotation data. For the specific process, please refer to Figure 1, which is the first schematic diagram of the application scenario of the method for processing aerial scene images provided in the present application. In the application scenario embodiment of the method for processing aerial scene images provided in the present application, the aerial scene image refers to an image containing aerial equipment and aerial-related equipment, such as: an image of an apron, an image of an airplane or drone flying, and an image related to airport operations, etc.; an aerial scene simulation image is a simulated image obtained by simulating a real aerial scene image based on computer vision processing; a real aerial scene image is a real image containing aerial equipment and aerial-related equipment that is pre-captured by an image acquisition device. In the application scenario of the method for processing aerial scene images provided in the present application, the application scenario of the method for processing aerial scene images provided in the present application is specifically described in detail by taking the apron scene image as an example. When the aerial scene image is other than the apron scene image, please refer to the application scenario of the method for processing the apron scene image, which will not be described one by one here. In practical applications, in order to more clearly introduce and display the image content to relevant personnel, or to facilitate computer equipment to more effectively identify and process images, relevant staff often need to annotate these images to obtain images with annotated data to assist people or computer equipment in understanding. For example, when training relevant network models for image processing, it is often necessary to annotate images used to participate in model training so that computer equipment can better understand them. Among them, in the application scenario embodiment of the method for processing aerial scene images provided in the present application, the annotated data generally includes one or more of the position data of the target object in the aerial scene image in the aerial scene image, the name data of the target object in the aerial scene image, and the contour data of the target object. Specifically, the position data of the target object in the aerial scene image in the aerial scene image generally corresponds to the area frame data of the target object in the aerial scene image.

[0145] If real aerial scene images are used, because real aerial images are images acquired by image acquisition equipment, when annotating data, it is generally necessary to manually or artificially annotate the data for the real aerial scene images. This is not only cumbersome to operate, but also prone to errors. By using image simulation and in the process of generating a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image, automatic data annotation of the target object in the target aerial scene simulation image is achieved. This can avoid consuming a lot of human resources and obtain an aerial scene simulation image with annotated data and a high degree of authenticity.

[0146] The method for processing aerial scene images provided in the present application is executed by a program or software for processing aerial scene images installed on an electronic device capable of providing computing or application services, such as a program or software for processing aerial scene images installed on a server. Specifically, the specific implementation process corresponding to the application scenario of the method for processing aerial scene images provided in the present application is as follows:

[0147] First, a target real apron scene image is obtained, and a simulated apron scene image to be processed is obtained. That is, the target real apron scene image to be simulated is first determined, and image simulation is performed on the target real apron scene image to obtain a simulated apron scene image to be processed corresponding to the target real apron scene image.

[0148] In the application scenario of the method for processing aviation scene images provided in this application, the specific operation of obtaining the simulated image of the apron scene to be processed is:

[0149] First, obtain the target object in the target real apron scene image and obtain the parameter data of the specified image acquisition device. The target object is the apron equipment and apron related equipment, such as apron aircraft, drones, buildings in the airport, airport facilities, etc.

[0150] Second, according to the target real apron scene image and the target object, a three-dimensional scene model corresponding to the target real apron scene image is constructed. The specific process is as follows: first, a target two-dimensional image of a three-dimensional object corresponding to the target object is obtained, that is, according to the target real apron scene image, a target two-dimensional image of a three-dimensional object corresponding to the target object is obtained by a designated image acquisition device. Secondly, the image feature data of the target two-dimensional image is extracted. That is, the feature point data, feature line data and feature area data of the target two-dimensional image are extracted. Thirdly, according to the image feature data of the target two-dimensional image, the corresponding relationship of the target object in the three-dimensional space coordinate system is determined. That is, according to the image feature data of the target two-dimensional image, the imaging point of the target object in the three-dimensional coordinate system is determined. Finally, according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system, a three-dimensional scene model is constructed in the target three-dimensional space coordinate system. That is, according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system, sparse reconstruction, dense reconstruction, point cloud modeling and texture mapping are further performed to construct the three-dimensional scene model.

[0151] It should be noted that in the application scenarios of the aerial scene image processing method provided in the present application, the specific processes corresponding to the sparse reconstruction, dense reconstruction, point cloud modeling and texture mapping processing are: according to the parameter data of the specified image acquisition device and the target three-dimensional space coordinate system, the sparse point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system is determined; the sparse point cloud data is expanded to obtain the dense point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system; based on the dense point cloud data, the three-dimensional scene model is constructed.

[0152] Third, the three-dimensional scene model corresponding to the apron scene simulation image to be processed is subjected to imaging simulation to obtain the apron scene simulation image to be processed. That is, according to the parameter data of the designated image acquisition device and the preset illumination parameter data, the three-dimensional scene model corresponding to the apron scene simulation image to be processed is subjected to imaging simulation to obtain the apron scene simulation image to be processed.

[0153] Secondly, obtain the annotation data of the objects in the apron scene simulation image to be processed. Since the application scenario of the aviation scene image processing method provided in the present application is to sequentially construct a three-dimensional scene model corresponding to the target real apron scene image through steps such as image acquisition, image feature extraction, image feature matching, sparse reconstruction, dense reconstruction, point cloud modeling, and texture mapping, and further perform imaging simulation on the three-dimensional scene model corresponding to the apron scene simulation image to be processed according to the parameter data of the specified image acquisition device and the preset illumination parameter data, and obtain the apron scene simulation image to be processed, therefore, after obtaining the apron scene simulation image to be processed, the corresponding annotation data of the objects in the apron scene simulation image to be processed can be determined according to the three-dimensional scene model, the parameter data of the specified image acquisition device, and the preset illumination parameter data.

[0154] Again, based on the apron scene simulation image to be processed and the target real apron scene image, a target apron scene simulation image is obtained. In the application scenario of the aviation scene image processing method provided in the present application, based on the apron scene simulation image to be processed and the target real apron scene image, the process of obtaining the target apron scene simulation image is as follows: the apron scene simulation image to be processed and the target real apron scene image are input into the target generative adversarial network model to obtain the target apron scene simulation image whose image feature data matches the image feature data of the target real apron scene image.

[0155] Because it is necessary to obtain the target apron scene simulation image through the target generative adversarial network model, in the application scenario of the aviation scene image processing method provided in this application, model training is also required to obtain the target generative adversarial network model. For details, please refer to Figure 2, which is a second schematic diagram of an application scenario of the method for processing aerial scene images provided in this application.

[0156] In the application scenario of the method for processing aviation scene images provided in this application, the principle of model training is: first, obtain a sample apron scene simulation image and a sample real apron scene image. Then, according to the sample apron scene simulation image and the sample real apron scene image, perform model training on the to-be-trained generative adversarial network model to obtain a target generative adversarial network model. The specific steps are as follows:

[0157] Step 1: Input the first sample apron scene simulation image and the first sample real apron scene image into the generative adversarial network model to be trained, and obtain the image feature data of the first sample real apron scene image; transfer the image feature data of the first sample real apron scene image to the first sample apron scene simulation image, and obtain the first target apron scene simulation image; judge whether the image feature data of the first target apron scene simulation image matches the image feature data of the first sample real apron scene image; if so, use the generative adversarial network model to be trained as the target generative adversarial network model. Wherein, obtaining the target apron scene simulation image is realized by the generator in the generative adversarial network model to be trained, that is, inputting the sample apron scene simulation image and the sample real apron scene image into the generator to generate the target apron scene simulation image; judging whether the image feature data of the target apron scene simulation image matches the image feature data of the sample real apron scene image is realized by the discriminator in the generative adversarial network model to be trained.

[0158] Step 2: If the image feature data of the first target apron scene simulation image does not match the image feature data of the first sample real apron scene image, the second sample apron scene simulation image and the second sample real apron scene image are input into the generative adversarial network model to be trained to obtain the image feature data of the second sample real apron scene image; the image feature data of the second sample real apron scene image is migrated to the second sample apron scene simulation image to obtain the second target apron scene simulation image; it is determined whether the image feature data of the second target apron scene simulation image matches the image feature data of the second sample real apron scene image; if the image feature data of the second target apron scene simulation image matches the image feature data of the second sample real apron scene image, the generative adversarial network model to be trained is used as the target generative adversarial network model.

[0159] Step 3: If the image feature data of the second target apron scene simulation image does not match the image feature data of the second sample real apron scene image, the sample apron scene simulation image and the sample real apron scene image are sequentially input into the to-be-trained generative adversarial network model until the target generative adversarial network model is obtained. That is, the to-be-trained generative adversarial network model can be used as the target generative adversarial network model only when the discriminator outputs a yes result.

[0160] In the application scenario of the method for processing aviation scene images provided in the present application, the process of obtaining a target apron scene simulation image whose image feature data matches the image feature data of the target real apron scene image is as follows: first, the apron scene simulation image to be processed and the target real apron scene image are input into the target generative adversarial network model to obtain high-dimensional image feature data of the target real apron scene image. Then, the high-dimensional image feature data of the target real apron scene image is migrated into the apron scene simulation image to be processed to obtain the target apron scene simulation image. Among them, the high-dimensional image feature data of the target real apron scene image includes at least one or more of image texture feature data, image clarity feature data, image color saturation feature data, image contrast feature data and image brightness data.

[0161] In the application scenario of the aerial scene image processing method provided in the present application, whether it is determined during the model training process or when using the target generative adversarial network model to generate the target apron scene simulation image, the similarity of image features between different images is judged. The principle of judging the similarity of image features between different images in the application scenario of the aerial scene image processing method provided in the present application is consistent. The following is a detailed explanation based on the judgment of the similarity of image features involved in obtaining the target apron scene simulation image.

[0162] The process of inputting the simulated apron scene image to be processed and the target real apron scene image into the target generative adversarial network model to obtain the target apron scene simulated image whose image feature data matches the image feature data of the target real apron scene image is as follows: first, inputting the simulated apron scene image to be processed and the target real apron scene image into the target generative adversarial network model to obtain the high-dimensional image feature data of the target real apron scene image; then, obtaining the initial high-dimensional image feature data of the simulated apron scene image to be processed; finally, adjusting the initial high-dimensional image feature data of the simulated apron scene image to be processed according to the high-dimensional image feature data of the target real apron scene image, and obtaining the target apron scene simulated image. That is, firstly inputting the simulated apron scene image to be processed and the target real apron scene image into the target generative adversarial network model to obtain the high-dimensional image feature data of the target real apron scene image; then, migrating the high-dimensional image feature data of the target real apron scene image into the simulated apron scene image to be processed, and obtaining the target apron scene simulated image.

[0163] When the initial high-dimensional image feature data of the apron scene simulation image to be processed is adjusted according to the high-dimensional image feature data of the target real apron scene image to obtain the target apron scene simulation image, the judgment of the similarity of image features is involved. The specific process is: according to the high-dimensional image feature data of the target real apron scene image, the initial high-dimensional image feature data of the apron scene simulation image to be processed is adjusted to obtain a candidate apron scene simulation image; the high-dimensional image feature data of the candidate apron scene simulation image is clustered and compared with the initial high-dimensional image feature data of the apron scene simulation image to be processed to obtain the similarity between the high-dimensional image feature data of the candidate apron scene simulation image and the initial high-dimensional image feature data of the apron scene simulation image to be processed; if the similarity reaches or exceeds the similarity threshold, the candidate apron scene simulation image is used as the target apron scene simulation image.

[0164] It should be noted that in the application scenario of the aerial scene image processing method provided in the present application, if the similarity does not reach the similarity threshold, the initial high-dimensional image feature data of the apron scene simulation image to be processed continues to be adjusted to obtain candidate apron scene simulation images until the target apron scene simulation image is obtained.

[0165] In the application scenario of the method for processing aviation scene images provided in the present application, after obtaining the target real apron scene image and the apron scene simulation image to be processed which is generated by image simulation and corresponds to the target real apron scene image, the apron scene simulation image to be processed and the target real apron scene image are input into the trained target generative adversarial network model to further obtain the target apron scene simulation image whose image feature data matches the image feature data of the target real apron scene image, thereby improving the realism of the apron scene simulation image.

[0166] Please refer to Figure 1 After obtaining the target apron scene simulation image and the annotation data of the object in the apron scene simulation image to be processed, the annotation data of the target object in the target apron scene simulation image can be obtained according to the annotation data of the object. That is, after obtaining the target apron scene simulation image according to the apron scene simulation image to be processed and the target real apron scene image, the annotation data of the target object in the target apron scene simulation image can be obtained according to the annotation data of the object further obtained after obtaining the apron scene simulation image to be processed that is generated by image simulation and corresponds to the target real apron scene image. In the application scenario of the method for processing aviation scene images provided in the present application, the steps for obtaining the annotation data of the target object in the target apron scene simulation image are: first, obtaining the matching relationship between the target object and the object. Then, according to the annotation data and the matching relationship of the object, the annotation data of the target object is obtained.

[0167] In the application scenario of the method for processing aerial scene images provided in the present application, after obtaining the apron scene simulation image to be processed generated by image simulation and corresponding to the target real apron scene image, the annotation data of the objects in the apron scene simulation image to be processed will be further obtained. Then, after obtaining the target apron scene simulation image based on the apron scene simulation image to be processed and the target real apron scene image, the annotation data of the target objects in the target apron scene simulation image can be further obtained based on the annotation data of the objects, thereby ensuring that the obtained target apron scene simulation image can carry the annotation data of the target objects in the target apron scene simulation image, thereby realizing automatic data annotation of the target objects in the target apron scene simulation image and simplifying the process of data annotation of the apron scene image.

[0168] It should be noted that the application scenario embodiment of the above-mentioned aerial scene image processing method provided in the present application is only an embodiment of the application scenario of the aerial scene image processing method provided in the present application. The purpose of providing the above-mentioned application scenario embodiment is to facilitate the understanding of the aerial scene image processing method provided in the present application, and is not used to limit the aerial scene image processing method provided in the present application. The aerial scene image processing method provided in the present application can also be applied to other application scenarios, and the aerial scene image processing method provided in the present application does not specifically limit the execution subject. When applying the aerial scene image processing method provided in the present application to other application scenarios, please refer to the application scenario embodiment of the above-mentioned aerial scene image processing method provided in the present application, which will not be repeated here one by one.

[0169] First embodiment

[0170] In the first embodiment of the present application, a method for processing an aerial scene image is provided. Figure 3-Figure 4 Provide explanation.

[0171] In step S301, a target real aerial scene image is obtained, and a simulated aerial scene image to be processed is obtained, where the simulated aerial scene image to be processed is a simulated image corresponding to the target real aerial scene image.

[0172] The so-called simulated aviation scene image to be processed is a simulated image corresponding to the target real aviation scene image, including: the simulated aviation scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aviation scene image.

[0173] In the first embodiment of the present application, an aviation scene image refers to an image containing aviation equipment and aviation-related equipment, such as an image of an aviation apron, an image of an airplane or drone flying, and images related to airport operations; an aviation scene simulation image is a simulated image obtained by simulating a real aviation scene image based on computer vision processing; a real aviation scene image is a real image that is pre-captured by an image acquisition device and contains aviation equipment and aviation-related equipment.

[0174] In the first embodiment of the present application, the sequence of obtaining the target real aerial scene image and obtaining the aerial scene simulation image to be processed is to first obtain the target real aerial scene image, and then further obtain the aerial scene simulation image to be processed based on the target real aerial scene image. Specifically, first, the target object in the target real aerial scene image is obtained; then, based on the target real aerial scene image and the target object, a three-dimensional scene model corresponding to the target real aerial scene image is constructed; finally, imaging simulation is performed on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed to obtain the aerial scene simulation image to be processed.

[0175] In the first embodiment of the present application, the method used to construct the three-dimensional scene model corresponding to the target real aerial scene image is generally three-dimensional reconstruction, that is, the two-dimensional image of the three-dimensional object corresponding to the target object, the parameter data of the image acquisition equipment are calibrated, the image feature data of the two-dimensional image is extracted, the target object is matched in the three-dimensional coordinate system, sparse reconstruction, dense reconstruction, point cloud modeling and texture mapping are performed in sequence to perform image simulation to achieve three-dimensional construction.

[0176] The detailed process of acquiring the two-dimensional image of the target object corresponding to the three-dimensional object is: obtaining the target object in the target real apron scene image, and obtaining the parameter data of the specified image acquisition device. Among them, the target object is aviation equipment and aviation-related equipment. When the aviation scene image is an apron scene image, the target object is generally: apron equipment and apron-related equipment, such as: apron aircraft, drones, buildings in the airport, airport facilities, etc.

[0177] According to the target real aerial scene image and the target object, the specific process of constructing the three-dimensional scene model corresponding to the target real aerial scene image is as follows: first, the target two-dimensional image of the three-dimensional object corresponding to the target object is obtained, that is, according to the target real aerial scene image, the target two-dimensional image of the three-dimensional object corresponding to the target object is obtained by a designated image acquisition device. Secondly, the image feature data of the target two-dimensional image is extracted. That is, the feature point data, feature line data and feature area data of the target two-dimensional image are extracted. Thirdly, according to the image feature data of the target two-dimensional image, the corresponding relationship of the target object in the three-dimensional space coordinate system is determined. That is, according to the image feature data of the target two-dimensional image, the imaging point of the target object in the three-dimensional coordinate system is determined. Finally, the parameter data of the designated image acquisition device is obtained, and according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system, the three-dimensional scene model is constructed in the target three-dimensional space coordinate system. That is, according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system, sparse reconstruction, dense reconstruction, point cloud modeling and texture mapping and other processing are further performed to construct the three-dimensional scene model.

[0178] It should be noted that in the first embodiment of the present application, the specific processes corresponding to the sparse reconstruction, dense reconstruction, point cloud modeling and texture mapping processing are: according to the parameter data of the specified image acquisition device and the target three-dimensional space coordinate system, the sparse point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system is determined; the sparse point cloud data is expanded to obtain the dense point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system; based on the dense point cloud data, the three-dimensional scene model is constructed.

[0179] The specific process of performing imaging simulation on the three-dimensional scene model corresponding to the apron scene simulation image to be processed and obtaining the apron scene simulation image to be processed is as follows: obtaining parameter data of a specified image acquisition device, performing imaging simulation on the three-dimensional scene model corresponding to the apron scene simulation image to be processed according to the parameter data of the specified image acquisition device and preset lighting parameter data, and obtaining the apron scene simulation image to be processed.

[0180] In step S302, the to-be-processed aviation scene simulation image and the target real aviation scene image are input into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0181] In the first embodiment of the present application, the target generative adversarial network model is a model pre-obtained through model training, and is specifically used to obtain image feature data of the target real aerial scene image based on the aerial scene simulation image to be processed and the target real aerial scene image, and migrate the image feature data of the target real aerial scene image to the aerial scene simulation image to be processed to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image.

[0182] The process of obtaining the target generative adversarial network model through model training is as follows: first, obtain a sample aerial scene simulation image and a sample real aerial scene image. Then, based on the sample aerial scene simulation image and the sample real aerial scene image, perform model training on the to-be-trained generative adversarial network model to obtain the target generative adversarial network model. The specific steps are as follows:

[0183] First, the first sample aviation scene simulation image and the first sample real aviation scene image are input into the generative adversarial network model to be trained to obtain the image feature data of the first sample real aviation scene image; the image feature data of the first sample real aviation scene image is transferred to the first sample aviation scene simulation image to obtain the first target aviation scene simulation image; it is judged whether the image feature data of the first target aviation scene simulation image matches the image feature data of the first sample real aviation scene image; if so, the generative adversarial network model to be trained is used as the target generative adversarial network model. Wherein, obtaining the target aviation scene simulation image is realized by the generator in the generative adversarial network model to be trained, that is, the sample aviation scene simulation image and the sample real aviation scene image are input into the generator to generate the target aviation scene simulation image; judging whether the image feature data of the target aviation scene simulation image matches the image feature data of the sample real aviation scene image is realized by the discriminator in the generative adversarial network model to be trained.

[0184] Then, if the image feature data of the first target aerial scene simulation image does not match the image feature data of the first sample real aerial scene image, the second sample aerial scene simulation image and the second sample real aerial scene image are input into the generative adversarial network model to be trained to obtain the image feature data of the second sample real aerial scene image; the image feature data of the second sample real aerial scene image is migrated to the second sample aerial scene simulation image to obtain the second target aerial scene simulation image; it is determined whether the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image; if the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image, the generative adversarial network model to be trained is used as the target generative adversarial network model.

[0185] Finally, if the image feature data of the second target aviation scene simulation image does not match the image feature data of the second sample real aviation scene image, the sample aviation scene simulation image and the sample real aviation scene image are sequentially input into the to-be-trained generative adversarial network model until the target generative adversarial network model is obtained. That is, only when the discriminator outputs a yes result, can the to-be-trained generative adversarial network model be used as the target generative adversarial network model.

[0186] In the first embodiment of the present application, the process of obtaining a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image is as follows: first, the aviation scene simulation image to be processed and the target real aviation scene image are input into the target generative adversarial network model to obtain high-dimensional image feature data of the target real aviation scene image. Then, the high-dimensional image feature data of the target real aviation scene image is transferred to the aviation scene simulation image to be processed to obtain the target aviation scene simulation image. Among them, the high-dimensional image feature data of the target real aviation scene image includes at least one or more of image texture feature data, image clarity feature data, image color saturation feature data, image contrast feature data and image brightness data.

[0187] In the first embodiment of the present application, whether it is determined during the model training process or when using the target generative adversarial network model to generate a target aviation scene simulation image, the similarity of image features between different images is judged. The principle of judging the similarity of image features between different images in the first embodiment of the present application is consistent. The following is a detailed explanation based on the judgment of the similarity of image features involved in obtaining the target aviation scene simulation image.

[0188] Please refer to the process of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into the target generative adversarial network model to obtain the target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image. Figure 4 , which is a flow chart of a method for generating a target aviation scene simulation image provided in the first embodiment of the present application. Step S401: Input the aviation scene simulation image to be processed and the target real aviation scene image into the target generative adversarial network model to obtain high-dimensional image feature data of the target real aviation scene image; Step S402: Obtain the initial high-dimensional image feature data of the aviation scene simulation image to be processed. Step S403: According to the high-dimensional image feature data of the target real aviation scene image, adjust the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain the target aviation scene simulation image. That is, first input the aviation scene simulation image to be processed and the target real aviation scene image into the target generative adversarial network model to obtain the high-dimensional image feature data of the target real aviation scene image; then migrate the high-dimensional image feature data of the target real aviation scene image into the aviation scene simulation image to be processed to obtain the target aviation scene simulation image.

[0189] When the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image is adjusted according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image, the judgment of the similarity of image features is involved. The specific process is: according to the high-dimensional image feature data of the target real aviation scene image, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image is adjusted to obtain a candidate aviation scene simulation image; the high-dimensional image feature data of the candidate aviation scene simulation image is clustered and compared with the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image; if the similarity reaches or exceeds the similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

[0190] It should be noted that in the first embodiment of the present application, if the similarity does not reach the similarity threshold, the initial high-dimensional image feature data of the aviation scene simulation image to be processed continues to be adjusted to obtain candidate aviation scene simulation images until the target aviation scene simulation image is obtained.

[0191] The method for processing aerial scene images provided in the first embodiment of the present application, after obtaining a target real aerial scene image and an aerial scene simulation image to be processed corresponding to the target real aerial scene image, inputs the aerial scene simulation image to be processed and the target real aerial scene image into a trained target generative adversarial network model, and further obtains a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image, thereby improving the realism of the aerial scene simulation image.

[0192] In the first embodiment of the present application, after obtaining the target aerial scene simulation image, the annotation data of the object in the aerial scene simulation image to be processed can be further obtained, and the matching relationship between the target object and the object in the target aerial scene simulation image can be obtained, and then the annotation data of the target object can be obtained according to the annotation data and the matching relationship of the object. Among them, the annotation data generally includes one or more of the position data of the target object in the aerial scene image in the aerial scene image, the name data of the target object in the aerial scene image, and the contour data of the target object. Specifically, the position data of the target object in the aerial scene image in the aerial scene image generally corresponds to the area frame data of the target object in the aerial scene image.

[0193] Obtaining the annotation data of the target object in the target real aerial scene image can more clearly introduce and display the image content to relevant personnel, or facilitate computer equipment to more accurately recognize and process the image.

[0194] Second embodiment

[0195] Corresponding to the application scenario embodiment of the method for processing aerial scene images provided by the present application and the method for processing aerial scene images provided by the first embodiment, the second embodiment of the present application also provides a device for processing aerial scene images. Since the device embodiment is basically similar to the application scenario embodiment and the first embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the application scenario embodiment and the first embodiment. The device embodiment described below is only exemplary.

[0196] Please refer to Figure 5 , which is a schematic diagram of an aerial scene image processing device provided in the second embodiment of the present application.

[0197] The aerial scene image processing device comprises:

[0198] The image acquisition unit 501 is used to acquire a target real aerial scene image and acquire a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0199] The image processing unit 502 is used to input the to-be-processed aviation scene simulation image and the target real aviation scene image into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0200] Optionally, the image processing unit 502 is specifically used to obtain a sample aerial scene simulation image and a sample real aerial scene image; based on the sample aerial scene simulation image and the sample real aerial scene image, the generative adversarial network model to be trained is trained to obtain the target generative adversarial network model.

[0201] Optionally, the performing model training on the to-be-trained generative adversarial network model according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model includes:

[0202] Inputting a first sample aviation scene simulation image and a first sample real aviation scene image into the to-be-trained generative adversarial network model to obtain image feature data of the first sample real aviation scene image;

[0203] Migrating the image feature data of the first sample real aviation scene image to the first sample aviation scene simulation image to obtain a first target aviation scene simulation image;

[0204] Determining whether the image feature data of the first target aviation scene simulation image matches the image feature data of the first sample real aviation scene image;

[0205] If so, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

[0206] Optionally, also include:

[0207] If the image feature data of the first target aviation scene simulation image does not match the image feature data of the first sample real aviation scene image, inputting the second sample aviation scene simulation image and the second sample real aviation scene image into the to-be-trained generative adversarial network model to obtain the image feature data of the second sample real aviation scene image;

[0208] Migrating the image feature data of the second sample real aviation scene image to the second sample aviation scene simulation image to obtain a second target aviation scene simulation image;

[0209] Determining whether the image feature data of the second target aviation scene simulation image matches the image feature data of the second sample real aviation scene image;

[0210] If the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

[0211] Optionally, it also includes: if the image feature data of the second target aerial scene simulation image does not match the image feature data of the second sample real aerial scene image, then inputting the sample aerial scene simulation image and the sample real aerial scene image into the generative adversarial network model to be trained in sequence until the target generative adversarial network model is obtained.

[0212] The optional image processing unit 502 is specifically used to input the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real aerial scene image; and migrate the high-dimensional image feature data of the target real aerial scene image into the to-be-processed aerial scene simulation image to obtain the target aerial scene simulation image.

[0213] Optionally, the step of transferring the high-dimensional image feature data of the target real aerial scene image into the to-be-processed aerial scene simulation image to obtain the target aerial scene simulation image includes:

[0214] Obtaining initial high-dimensional image feature data of the to-be-processed aerial scene simulation image;

[0215] According to the high-dimensional image feature data of the target real aviation scene image, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image is adjusted to obtain the target aviation scene simulation image.

[0216] Optionally, adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image includes:

[0217] According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain a candidate aviation scene simulation image;

[0218] Performing clustering comparison on the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed;

[0219] If the similarity reaches or exceeds a similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

[0220] Optionally, also include:

[0221] If the similarity does not reach the similarity threshold, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image continues to be adjusted to obtain candidate aviation scene simulation images until the target aviation scene simulation image is obtained.

[0222] Optionally, the high-dimensional image feature data of the target real aerial scene image includes at least one or more of image texture feature data, image clarity feature data, image color saturation feature data, image contrast feature data and image brightness data.

[0223] Optionally, the image acquisition unit 501 is specifically used to obtain the target object in the target real aerial scene image; construct a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object; perform imaging simulation on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed, and obtain the aerial scene simulation image to be processed.

[0224] Optionally, constructing a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object includes:

[0225] According to the target real aerial scene image, obtaining a target two-dimensional image of a three-dimensional object corresponding to the target object through a designated image acquisition device;

[0226] Establishing a target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device;

[0227] Extracting image feature data of the target two-dimensional image;

[0228] Determining a corresponding relationship of the target object in the three-dimensional space coordinate system according to the image feature data of the target two-dimensional image;

[0229] The three-dimensional scene model is constructed in the target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system.

[0230] Optionally, constructing the three-dimensional scene model in the target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system includes:

[0231] Determining, according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system, the sparse point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system;

[0232] Performing point cloud expansion on the sparse point cloud data to obtain dense point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system;

[0233] The three-dimensional scene model is obtained and constructed according to the dense point cloud data.

[0234] Optionally, performing imaging simulation on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed to obtain the aerial scene simulation image to be processed includes: performing imaging simulation on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed according to parameter data of a designated image acquisition device and preset lighting parameter data to obtain the aerial scene simulation image to be processed.

[0235] Optionally, the method further includes: obtaining parameter data of the designated image acquisition device.

[0236] Optionally, the aerial scene image processing device provided in the second embodiment of the present application further includes:

[0237] A first annotation data obtaining unit, used to obtain annotation data of objects in the aerial scene simulation image to be processed;

[0238] A matching relationship obtaining unit, used for obtaining a matching relationship between a target object in the target aviation scene simulation image and the object;

[0239] The second annotation data obtaining unit is used to obtain the annotation data of the target object according to the annotation data of the object and the matching relationship.

[0240] Optionally, the simulated aerial scene image to be processed is a simulated image corresponding to the target real aerial scene image, including: the simulated aerial scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aerial scene image.

[0241] Third embodiment

[0242] Corresponding to the application scenario embodiment of the method for processing aerial scene images provided in the present application and the method for processing aerial scene images provided in the first embodiment, the third embodiment of the present application further provides an electronic device.

[0243] As shown in 6, Figure 6 A schematic diagram of an electronic device provided in an embodiment of the present application.

[0244] The electronic device includes:

[0245] Processor 601; and

[0246] The memory 602 is used to store a program of a method for processing an aerial scene image. After the device is powered on and the program of the method for processing an aerial scene image is run by the processor, the following steps are performed:

[0247] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0248] The to-be-processed aviation scene simulation image and the target real aviation scene image are input into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0249] It should be noted that, for the detailed description of the electronic device provided in the third embodiment of the present application, reference can be made to the application scenario embodiment of the method for processing aerial scene images provided in the present application and the related description of the method for processing aerial scene images provided in the first embodiment, which will not be repeated here.

[0250] Fourth embodiment

[0251] Corresponding to the application scenario embodiment of the method for processing aerial scene images provided in the present application and the method for processing aerial scene images provided in the first embodiment, the fourth embodiment of the present application further provides a storage medium storing a program of the method for processing aerial scene images, the program being executed by a processor to perform the following steps:

[0252] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0253] The to-be-processed aviation scene simulation image and the target real aviation scene image are input into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0254] It should be noted that for the detailed description of the storage medium provided in the fourth embodiment of the present application, reference can be made to the application scenario embodiment of the aerial scene image processing method provided in the present application and the related description of the aerial scene image processing method provided in the first embodiment, and no further details will be given here.

[0255] Fifth embodiment

[0256] Corresponding to the scenario embodiment of the method for processing aerial scene images provided in the present application and the method for processing aerial scene images provided in the first embodiment, the fifth embodiment of the present application provides another method for processing aerial scene images. Since the method for processing aerial scene images in the fifth embodiment is basically similar to the application scenario embodiment and the first embodiment, the description is relatively simple, and the relevant parts can be referred to the application scenario embodiment and the partial description of the first embodiment. The fifth embodiment described below is merely illustrative.

[0257] Please refer to Figure 7 , which is a flowchart of a method for processing aerial scene images provided in the fifth embodiment of the present application.

[0258] Step S701: obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, where the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image.

[0259] The so-called simulated aviation scene image to be processed is a simulated image corresponding to the target real aviation scene image, including: the simulated aviation scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aviation scene image.

[0260] In the fifth embodiment of the present application, the aviation scene image refers to an image containing aviation equipment and aviation-related equipment, such as: an image of an aviation apron, an image of an airplane or drone flying, and images related to airport operations, etc.; the aviation scene simulation image is a simulated image obtained by simulating a real aviation scene image based on computer vision processing; the real aviation scene image is a real image that is pre-captured by an image acquisition device and contains aviation equipment and aviation-related equipment.

[0261] In the fifth embodiment of the present application, the process of obtaining a target real aerial scene image and obtaining a simulated aerial scene image to be processed is: first obtain the target real aerial scene image, and then further obtain the simulated aerial scene image to be processed based on the target real aerial scene image. Specifically, first, obtain the target object in the target real aerial scene image; then, construct a three-dimensional scene model corresponding to the target real aerial scene image based on the target real aerial scene image and the target object; finally, perform imaging simulation on the three-dimensional scene model corresponding to the simulated aerial scene image to be processed to obtain the simulated aerial scene image to be processed.

[0262] Step S702: Obtaining annotation data of objects in the aerial scene simulation image to be processed.

[0263] In a fifth embodiment of the present application, the annotation data generally includes one or more of the position data of the target object in the aerial scene image, the name data of the target object in the aerial scene image, and the contour data of the target object. Specifically, the position data of the target object in the aerial scene image generally corresponds to the area frame data of the target object in the aerial scene image. Since the aerial scene simulation image to be processed is generated by image simulation, after the image simulation is completed, the annotation data of the object in the aerial scene simulation image to be processed can be further obtained.

[0264] Step S703: obtaining a target aviation scene simulation image according to the aviation scene simulation image to be processed and the target real aviation scene image.

[0265] In the fifth embodiment of the present application, the operation of obtaining the target aviation scene simulation image according to the aviation scene simulation image to be processed and the target real aviation scene image is: inputting the aviation scene simulation image to be processed and the target real aviation scene image into the target generative adversarial network model to obtain the target aviation scene simulation image.

[0266] In the fifth embodiment of the present application, the target generative adversarial network model is a model pre-obtained through model training, and is specifically used to obtain image feature data of the target real aviation scene image based on the aviation scene simulation image to be processed and the target real aviation scene image, and migrate the image feature data of the target real aviation scene image to the aviation scene simulation image to be processed to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image.

[0267] The process of obtaining the target generative adversarial network model through model training is as follows: first, obtain a sample aerial scene simulation image and a sample real aerial scene image. Then, based on the sample aerial scene simulation image and the sample real aerial scene image, perform model training on the to-be-trained generative adversarial network model to obtain the target generative adversarial network model. The specific steps are as follows:

[0268] First, the first sample aviation scene simulation image and the first sample real aviation scene image are input into the generative adversarial network model to be trained to obtain the image feature data of the first sample real aviation scene image; the image feature data of the first sample real aviation scene image is transferred to the first sample aviation scene simulation image to obtain the first target aviation scene simulation image; it is judged whether the image feature data of the first target aviation scene simulation image matches the image feature data of the first sample real aviation scene image; if so, the generative adversarial network model to be trained is used as the target generative adversarial network model. Wherein, obtaining the target aviation scene simulation image is realized by the generator in the generative adversarial network model to be trained, that is, the sample aviation scene simulation image and the sample real aviation scene image are input into the generator to generate the target aviation scene simulation image; judging whether the image feature data of the target aviation scene simulation image matches the image feature data of the sample real aviation scene image is realized by the discriminator in the generative adversarial network model to be trained.

[0269] Then, if the image feature data of the first target aerial scene simulation image does not match the image feature data of the first sample real aerial scene image, the second sample aerial scene simulation image and the second sample real aerial scene image are input into the generative adversarial network model to be trained to obtain the image feature data of the second sample real aerial scene image; the image feature data of the second sample real aerial scene image is migrated to the second sample aerial scene simulation image to obtain the second target aerial scene simulation image; it is determined whether the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image; if the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image, the generative adversarial network model to be trained is used as the target generative adversarial network model.

[0270] Finally, if the image feature data of the second target aviation scene simulation image does not match the image feature data of the second sample real aviation scene image, the sample aviation scene simulation image and the sample real aviation scene image are sequentially input into the to-be-trained generative adversarial network model until the target generative adversarial network model is obtained. That is, only when the discriminator outputs a yes result, can the to-be-trained generative adversarial network model be used as the target generative adversarial network model.

[0271] In the fifth embodiment of the present application, the process of obtaining a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image is as follows: first, the aviation scene simulation image to be processed and the target real aviation scene image are input into the target generative adversarial network model to obtain high-dimensional image feature data of the target real aviation scene image. Then, the high-dimensional image feature data of the target real aviation scene image is transferred to the aviation scene simulation image to be processed to obtain the target aviation scene simulation image. Among them, the high-dimensional image feature data of the target real aviation scene image includes at least one or more of image texture feature data, image clarity feature data, image color saturation feature data, image contrast feature data and image brightness data.

[0272] In the fifth embodiment of the present application, whether it is determined during the model training process or when using the target generative adversarial network model to generate a target aviation scene simulation image, the similarity of image features between different images is judged. The principle of judging the similarity of image features between different images in the fifth embodiment of the present application is consistent. The following is a detailed explanation based on the judgment of the similarity of image features involved in obtaining the target aviation scene simulation image.

[0273] The process of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into the target generative adversarial network model to obtain the target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image is as follows: first, the to-be-processed aerial scene simulation image and the target real aerial scene image are input into the target generative adversarial network model to obtain the high-dimensional image feature data of the target real aerial scene image; then, the initial high-dimensional image feature data of the to-be-processed aerial scene simulation image is obtained. Finally, according to the high-dimensional image feature data of the target real aerial scene image, the initial high-dimensional image feature data of the to-be-processed aerial scene simulation image is adjusted to obtain the target aerial scene simulation image. That is, the to-be-processed aerial scene simulation image and the target real aerial scene image are first input into the target generative adversarial network model to obtain the high-dimensional image feature data of the target real aerial scene image; then, the high-dimensional image feature data of the target real aerial scene image is transferred into the to-be-processed aerial scene simulation image to obtain the target aerial scene simulation image.

[0274] When the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image is adjusted according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image, the judgment of the similarity of image features is involved. The specific process is: according to the high-dimensional image feature data of the target real aviation scene image, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image is adjusted to obtain a candidate aviation scene simulation image; the high-dimensional image feature data of the candidate aviation scene simulation image is clustered and compared with the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image; if the similarity reaches or exceeds the similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

[0275] It should be noted that in the fifth embodiment of the present application, if the similarity does not reach the similarity threshold, the initial high-dimensional image feature data of the aviation scene simulation image to be processed continues to be adjusted to obtain candidate aviation scene simulation images until the target aviation scene simulation image is obtained.

[0276] Step S704: obtaining the annotation data of the target object in the target aerial scene simulation image according to the annotation data of the object.

[0277] After obtaining the target aviation scene simulation image and the annotation data of the object in the aviation scene simulation image to be processed, the annotation data of the target object in the target aviation scene simulation image can be obtained according to the annotation data of the object. That is, after obtaining the target aviation scene simulation image according to the aviation scene simulation image to be processed and the target real aviation scene image, the annotation data of the target object in the target aviation scene simulation image can be further obtained according to the annotation data of the object further obtained after obtaining the aviation scene simulation image to be processed corresponding to the target real aviation scene image. In the fifth embodiment, the steps of obtaining the annotation data of the target object in the target aviation scene simulation image are: first, obtaining the matching relationship between the target object and the object. Then, according to the annotation data of the object and the matching relationship, the annotation data of the target object is obtained.

[0278] The method for processing aerial scene images provided in the fifth embodiment, after obtaining the aerial scene simulation image to be processed corresponding to the target real aerial scene image, will further obtain the annotation data of the object in the aerial scene simulation image to be processed, and then after obtaining the target aerial scene simulation image based on the aerial scene simulation image to be processed and the target real aerial scene image, it can further obtain the annotation data of the target object in the target aerial scene simulation image based on the annotation data of the object, thereby ensuring that the obtained target aerial scene simulation image can carry the annotation data of the target object in the target aerial scene simulation image, thereby realizing automatic data annotation of the target object in the target aerial scene simulation image, and simplifying the process of data annotation for aerial scene images.

[0279] Sixth embodiment

[0280] Corresponding to the application scenario embodiment of the method for processing aerial scene images provided by the present application and the method for processing aerial scene images provided by the fifth embodiment, the sixth embodiment of the present application also provides a device for processing aerial scene images. Since the device embodiment is basically similar to the application scenario embodiment and the fifth embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the application scenario embodiment and the fifth embodiment. The device embodiment described below is only illustrative.

[0281] Please refer to Figure 8 , which is a schematic diagram of an aerial scene image processing device provided in the sixth embodiment of the present application.

[0282] The aerial scene image processing device comprises:

[0283] The image acquisition unit 801 is used to acquire a target real aerial scene image and acquire a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0284] The annotation data obtaining unit 802 is used to obtain the annotation data of the object in the aerial scene simulation image to be processed;

[0285] A target image acquisition unit 803 is used to acquire a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image;

[0286] The annotation data obtaining unit 804 is used to obtain the annotation data of the target object in the target aerial scene simulation image according to the annotation data of the object.

[0287] The annotated data obtaining unit 804 is specifically configured to obtain a matching relationship between the target object and the object; and obtain the annotated data of the target object according to the annotated data of the object and the matching relationship.

[0288] Optionally, the annotation data of the object includes at least one or more of the position data of the object in the aerial scene simulation image to be processed, the name data of the object and the contour data of the object.

[0289] Optionally, the target image acquisition unit 803 specifically inputs the to-be-processed aviation scene simulation image and the target real aviation scene image into a target generative adversarial network model to obtain the target aviation scene simulation image.

[0290] Optionally, the step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises:

[0291] Obtaining a sample aerial scene simulation image and obtaining a sample real aerial scene image;

[0292] The to-be-trained generative adversarial network model is trained according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model.

[0293] Optionally, the performing model training on the to-be-trained generative adversarial network model according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model includes:

[0294] Inputting a first sample aviation scene simulation image and a first sample real aviation scene image into the to-be-trained generative adversarial network model to obtain image feature data of the first sample real aviation scene image;

[0295] Migrating the image feature data of the first sample real aviation scene image to the first sample aviation scene simulation image to obtain a first target aviation scene simulation image;

[0296] Determining whether the image feature data of the first target aviation scene simulation image matches the image feature data of the first sample real aviation scene image;

[0297] If so, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

[0298] Optionally, also include:

[0299] If the image feature data of the first target aviation scene simulation image does not match the image feature data of the first sample real aviation scene image, inputting the second sample aviation scene simulation image and the second sample real aviation scene image into the to-be-trained generative adversarial network model to obtain the image feature data of the second sample real aviation scene image;

[0300] Migrating the image feature data of the second sample real aviation scene image to the second sample aviation scene simulation image to obtain a second target aviation scene simulation image;

[0301] Determining whether the image feature data of the second target aviation scene simulation image matches the image feature data of the second sample real aviation scene image;

[0302] If the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

[0303] Optionally, it also includes: if the image feature data of the second target aerial scene simulation image does not match the image feature data of the second sample real aerial scene image, then inputting the sample aerial scene simulation image and the sample real aerial scene image into the generative adversarial network model to be trained in sequence until the target generative adversarial network model is obtained.

[0304] Optionally, the to-be-processed aerial scene simulation image and the target real aerial scene image are input into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image, including:

[0305] Inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real aerial scene image;

[0306] The high-dimensional image feature data of the target real aviation scene image is transferred into the aviation scene simulation image to be processed to obtain the target aviation scene simulation image.

[0307] Optionally, the step of transferring the high-dimensional image feature data of the target real aerial scene image into the to-be-processed aerial scene simulation image to obtain the target aerial scene simulation image includes:

[0308] Obtaining initial high-dimensional image feature data of the to-be-processed aerial scene simulation image;

[0309] According to the high-dimensional image feature data of the target real aviation scene image, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image is adjusted to obtain the target aviation scene simulation image.

[0310] Optionally, adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image includes:

[0311] According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain a candidate aviation scene simulation image;

[0312] Performing clustering comparison on the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed;

[0313] If the similarity reaches or exceeds a similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

[0314] Optionally, the image acquisition unit 801 is specifically used to obtain the target object in the target real aerial scene image; construct a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object; perform imaging simulation on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed to obtain the aerial scene simulation image to be processed; obtaining the annotation data of the object in the aerial scene simulation image to be processed includes: obtaining the annotation data of the object according to the three-dimensional scene model corresponding to the aerial scene simulation image to be processed.

[0315] Optionally, the simulated aerial scene image to be processed is a simulated image corresponding to the target real aerial scene image, including: the simulated aerial scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aerial scene image.

[0316] Seventh embodiment

[0317] Corresponding to the application scenario embodiment of the method for processing aerial scene images provided in the present application and the method for processing aerial scene images provided in the fifth embodiment, the seventh embodiment of the present application further provides an electronic device.

[0318] As shown in 6, Figure 6A schematic diagram of an electronic device provided in an embodiment of the present application.

[0319] The electronic device includes:

[0320] Processor 601; and

[0321] The memory 602 is used to store a program of a method for processing an aerial scene image. After the device is powered on and the program of the method for processing an aerial scene image is run by the processor, the following steps are performed:

[0322] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0323] Obtaining annotation data of objects in the aerial scene simulation image to be processed;

[0324] Obtaining a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image;

[0325] The labeling data of the target object in the target aerial scene simulation image is obtained according to the labeling data of the object.

[0326] It should be noted that, for the detailed description of the electronic device provided in the seventh embodiment of the present application, reference can be made to the application scenario embodiment of the aerial scene image processing method provided in the present application and the related description of the aerial scene image processing method provided in the fifth embodiment, which will not be repeated here.

[0327] Eighth embodiment

[0328] Corresponding to the application scenario embodiment of the method for processing aerial scene images provided in the present application and the method for processing aerial scene images provided in the fifth embodiment, the eighth embodiment of the present application further provides a storage medium storing a program of the method for processing aerial scene images, the program being executed by a processor to perform the following steps:

[0329] Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image;

[0330] Obtaining annotation data of objects in the aerial scene simulation image to be processed;

[0331] Obtaining a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image;

[0332] The labeling data of the target object in the target aerial scene simulation image is obtained according to the labeling data of the object.

[0333] It should be noted that for the detailed description of the storage medium provided in the eighth embodiment of the present application, reference can be made to the application scenario embodiment of the aerial scene image processing method provided in the present application and the related description of the aerial scene image processing method provided in the fifth embodiment, and no further details will be given here.

[0334] Ninth embodiment

[0335] Corresponding to the scenario embodiment of the method for processing aerial scene images provided in the present application and the method for processing aerial scene images provided in the first embodiment, the ninth embodiment of the present application provides another method for processing aerial scene images. Since the method for processing aerial scene images in the ninth embodiment is basically similar to the application scenario embodiment and the first embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the application scenario embodiment and the first embodiment. The ninth embodiment described below is merely illustrative.

[0336] The steps corresponding to the image processing method provided in the ninth embodiment of the present application are as follows:

[0337] First, a target real image is obtained, and a simulated image to be processed is obtained, where the simulated image to be processed is a simulated image corresponding to the target real image.

[0338] The so-called simulated aviation scene image to be processed is a simulated image corresponding to the target real aviation scene image, including: the simulated aviation scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aviation scene image.

[0339] The image processing method provided in the ninth embodiment of the present application is generally applied to aviation scenes, and specifically used to process aviation scene images. At this time, the target real image and the to-be-processed simulation image corresponding to the target real image are generally aviation scene images, wherein the aviation scene image refers to an image containing aviation equipment and aviation-related equipment, such as: an image of an aviation apron, an image of an airplane or drone flying, and airport operation-related images, etc.; the aviation scene simulation image is a simulated image obtained by performing image simulation on the real aviation scene image based on computer vision processing; the real aviation scene image is a real image pre-captured by an image acquisition device, containing aviation equipment and aviation-related equipment.

[0340] In addition, the image processing method provided in the ninth embodiment of the present application can also be applied to urban traffic management, urban management, industrial workshop safety management, construction site safety management, and highway traffic management, etc.

[0341] When the image processing method provided in the ninth embodiment of the present application is applied to urban traffic management, the target real image and the simulated image to be processed corresponding to the target real image are generally urban road scene images, wherein the urban road scene image refers to an image containing urban roads, road equipment and vehicles, such as: images of vehicle operation on urban roads, public transportation or pedestrian related images, etc.; the urban road scene simulation image is a simulated image obtained by simulating the real urban road scene image based on computer vision processing; the real urban road scene image is a real image of urban roads, road equipment and vehicles, which is pre-captured by an image acquisition device.

[0342] When the image processing method provided in the ninth embodiment of the present application is applied to the safety management of industrial workshops, the target real image and the to-be-processed simulated image corresponding to the target real image are generally industrial workshop scene images, wherein the industrial workshop scene image refers to an image containing industrial plants, industrial production equipment, and workshop staff, such as: workshop assembly line operation images, workshop equipment operation, or staff communication related images, etc.; the industrial workshop scene simulation image is a simulated image obtained by performing image simulation on the real industrial workshop scene image based on computer vision processing; the real industrial workshop scene image is a real image containing industrial plants, industrial production equipment, and workshop staff that is pre-collected by an image acquisition device. Then, the to-be-processed simulated image and the target real image are input into the target generative adversarial network model to obtain a target simulation image whose image feature data matches the image feature data of the target real image.

[0343] In the ninth embodiment of the present application, the target generative adversarial network model is a model pre-obtained through model training, and is specifically used to obtain image feature data of the target real image based on the simulated image to be processed and the target real image, and migrate the image feature data of the target real image to the simulated image to be processed to obtain a target simulated image whose image feature data matches the image feature data of the target real image.

[0344] Tenth embodiment

[0345] Corresponding to the scenario embodiment of the method for processing aerial scene images provided in the present application and the method for processing aerial scene images provided in the fifth embodiment, the tenth embodiment of the present application provides another method for processing aerial scene images. Since the method for processing aerial scene images in the tenth embodiment is basically similar to the application scenario embodiment and the fifth embodiment, the description is relatively simple, and the relevant parts refer to the partial description of the application scenario embodiment and the fifth embodiment. The tenth embodiment described below is merely illustrative.

[0346] The so-called simulated aviation scene image to be processed is a simulated image corresponding to the target real aviation scene image, including: the simulated aviation scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aviation scene image.

[0347] The steps corresponding to the image processing method provided in the tenth embodiment of the present application are as follows:

[0348] First, a target real image is obtained, and a simulated image to be processed is obtained, where the simulated image to be processed is a simulated image corresponding to the target real image.

[0349] The image processing method provided in the tenth embodiment of the present application is generally applied to aviation scenes, and specifically used to process aviation scene images. At this time, the target real image and the to-be-processed simulation image corresponding to the target real image are generally aviation scene images, wherein the aviation scene image refers to an image containing aviation equipment and aviation-related equipment, such as: an image of an aviation apron, an image of an airplane or drone flying, and images related to airport operations, etc.; the aviation scene simulation image is a simulated image obtained by performing image simulation on the real aviation scene image based on computer vision processing; the real aviation scene image is a real image pre-captured by an image acquisition device, containing aviation equipment and aviation-related equipment.

[0350] In addition, the image processing method provided in the tenth embodiment of the present application can also be applied to urban traffic management, urban management, industrial workshop safety management, construction site safety management, and highway traffic management, etc.

[0351] When the image processing method provided in the tenth embodiment of the present application is applied to urban traffic management, the target real image and the simulated image to be processed corresponding to the target real image are generally urban road scene images, wherein the urban road scene image refers to an image containing urban roads, road equipment and vehicles, such as: images of vehicle operation on urban roads, public transportation or pedestrian related images, etc.; the urban road scene simulation image is a simulated image obtained by performing image simulation on the real urban road scene image based on computer vision processing; the real urban road scene image is a real image of urban roads, road equipment and vehicles, which is pre-captured by an image acquisition device.

[0352] When the image processing method provided in the tenth embodiment of the present application is applied to industrial workshop safety management, the target real image and the simulated image to be processed corresponding to the target real image are generally industrial workshop scene images, wherein the industrial workshop scene image refers to an image containing industrial plant, industrial production equipment and workshop staff, such as: workshop assembly line operation images, workshop equipment operation or staff communication related images, etc.; the industrial workshop scene simulation image is a simulated image obtained by performing image simulation of the real industrial workshop scene image based on computer vision processing; the real industrial workshop scene image is a real image pre-captured by an image acquisition device, containing industrial plant, industrial production equipment and workshop staff.

[0353] The so-called simulated aviation scene image to be processed is a simulated image corresponding to the target real aviation scene image, including: the simulated aviation scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aviation scene image.

[0354] Secondly, obtain the annotation data of the object in the simulated image to be processed.

[0355] In the tenth embodiment of the present application, the annotation data generally includes one or more of position data of a target object in an image, name data of the target object in the image, and contour data of the target object. Specifically, the position data of the target object in the image generally corresponds to the area box data of the target object in the image.

[0356] Thirdly, a target simulated image is obtained according to the simulated image to be processed and the target real image.

[0357] In the tenth embodiment of the present application, the operation of obtaining the target simulated image according to the simulated image to be processed and the target real image is: inputting the simulated image to be processed and the target real image into the target generative adversarial network model to obtain the target simulated image.

[0358] In the tenth embodiment of the present application, the target generative adversarial network model is a model pre-obtained through model training, and is specifically used to obtain image feature data of the target real image based on the simulated image to be processed and the target real image, and migrate the image feature data of the target real image to the simulated image to be processed to obtain a target simulated image whose image feature data matches the image feature data of the target real image.

[0359] Finally, the annotation data of the target object in the target simulation image is obtained according to the annotation data of the object.

[0360] After obtaining the target simulated image and the annotation data of the object in the simulated image to be processed, the annotation data of the target object in the target simulated image can be obtained according to the annotation data of the object. That is, after obtaining the target simulated image according to the simulated image to be processed and the target real image, the annotation data of the target object in the target simulated image can be further obtained according to the annotation data of the object further obtained after obtaining the simulated image to be processed corresponding to the target real image. In the fifth embodiment, the steps of obtaining the annotation data of the target object in the target simulated image are: first, obtaining the matching relationship between the target object and the object. Then, according to the annotation data of the object and the matching relationship, the annotation data of the target object is obtained.

[0361] Although the present application is disclosed as above in the preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

[0362] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0363] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0364] 1. Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer-readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0365] 2. Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

Claims

1. A method for processing an aerial scene image, characterized in that: include: Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image; Inputting the to-be-processed aviation scene simulation image and the target real aviation scene image into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image; The step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises: Inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real aerial scene image; Migrating the high-dimensional image feature data of the target real aviation scene image into the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image; The step of transferring the high-dimensional image feature data of the target real aviation scene image into the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image comprises: Obtaining initial high-dimensional image feature data of the to-be-processed aerial scene simulation image; According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image; The step of adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image comprises: According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain a candidate aviation scene simulation image; Performing clustering comparison on the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed; If the similarity reaches or exceeds a similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

2. The method for processing aerial scene images according to claim 1, characterized in that: The step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises: Obtaining a sample aerial scene simulation image and obtaining a sample real aerial scene image; The to-be-trained generative adversarial network model is trained according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model.

3. The method for processing aerial scene images according to claim 2, characterized in that: The method of performing model training on the to-be-trained generative adversarial network model according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model includes: Inputting a first sample aviation scene simulation image and a first sample real aviation scene image into the to-be-trained generative adversarial network model to obtain image feature data of the first sample real aviation scene image; Migrating the image feature data of the first sample real aviation scene image to the first sample aviation scene simulation image to obtain a first target aviation scene simulation image; Determining whether the image feature data of the first target aviation scene simulation image matches the image feature data of the first sample real aviation scene image; If so, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

4. The method for processing aerial scene images according to claim 3, characterized in that: Also includes: If the image feature data of the first target aviation scene simulation image does not match the image feature data of the first sample real aviation scene image, inputting the second sample aviation scene simulation image and the second sample real aviation scene image into the to-be-trained generative adversarial network model to obtain the image feature data of the second sample real aviation scene image; Migrating the image feature data of the second sample real aviation scene image to the second sample aviation scene simulation image to obtain a second target aviation scene simulation image; Determining whether the image feature data of the second target aviation scene simulation image matches the image feature data of the second sample real aviation scene image; If the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

5. The method for processing aerial scene images according to claim 4, characterized in that: Also includes: If the image feature data of the second target aerial scene simulation image does not match the image feature data of the second sample real aerial scene image, the sample aerial scene simulation image and the sample real aerial scene image are sequentially input into the generative adversarial network model to be trained until the target generative adversarial network model is obtained.

6. The method for processing aerial scene images according to claim 1, characterized in that: Also includes: If the similarity does not reach the similarity threshold, the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image continues to be adjusted to obtain candidate aviation scene simulation images until the target aviation scene simulation image is obtained.

7. The method for processing aerial scene images according to claim 1, characterized in that: The high-dimensional image feature data of the target real aerial scene image includes at least one or more of image texture feature data, image clarity feature data, image color saturation feature data, image contrast feature data and image brightness data.

8. The method for processing aerial scene images according to claim 1, characterized in that: The step of obtaining a target real aerial scene image and obtaining a simulated aerial scene image to be processed includes: Obtaining a target object in the target real aerial scene image; Constructing a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object; Perform imaging simulation on the three-dimensional scene model corresponding to the aviation scene simulation image to be processed to obtain the aviation scene simulation image to be processed.

9. The method for processing aerial scene images according to claim 8, characterized in that: The step of constructing a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object comprises: According to the target real aerial scene image, obtaining a target two-dimensional image of a three-dimensional object corresponding to the target object through a designated image acquisition device; Establishing a target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device; Extracting image feature data of the target two-dimensional image; Determining a corresponding relationship of the target object in the three-dimensional space coordinate system according to the image feature data of the target two-dimensional image; The three-dimensional scene model is constructed in the target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system.

10. The method for processing aerial scene images according to claim 9, characterized in that: The step of constructing the three-dimensional scene model in the target three-dimensional space coordinate system according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system comprises: Determining, according to the parameter data of the designated image acquisition device and the target three-dimensional space coordinate system, the sparse point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system; Performing point cloud expansion on the sparse point cloud data to obtain dense point cloud data corresponding to the three-dimensional scene model in the target three-dimensional space coordinate system; The three-dimensional scene model is obtained and constructed according to the dense point cloud data.

11. The method for processing aerial scene images according to claim 8, characterized in that: The performing imaging simulation on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed to obtain the aerial scene simulation image to be processed includes: performing imaging simulation on the three-dimensional scene model corresponding to the aerial scene simulation image to be processed according to parameter data of a designated image acquisition device and preset illumination parameter data to obtain the aerial scene simulation image to be processed.

12. The method for processing aerial scene images according to claim 9, characterized in that: Also includes: Obtain parameter data of the specified image acquisition device.

13. The method for processing aerial scene images according to claim 1, characterized in that: Also includes: Obtaining annotation data of objects in the aerial scene simulation image to be processed; Obtaining a matching relationship between a target object in the target aerial scene simulation image and the object; The labeled data of the target object is obtained according to the labeled data of the object and the matching relationship.

14. The method for processing aerial scene images according to claim 1, characterized in that: The simulated aviation scene image to be processed is a simulated image corresponding to the target real aviation scene image, including: the simulated aviation scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aviation scene image.

15. A device for processing aerial scene images, characterized in that: include: An image acquisition unit, used to acquire a target real aerial scene image and to acquire a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image; An image processing unit, used for inputting the to-be-processed aviation scene simulation image and the target real aviation scene image into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image; The image processing unit is further used for: Inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real aerial scene image; Migrating the high-dimensional image feature data of the target real aviation scene image into the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image; The step of transferring the high-dimensional image feature data of the target real aviation scene image into the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image comprises: Obtaining initial high-dimensional image feature data of the to-be-processed aerial scene simulation image; According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image; The step of adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image comprises: According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain a candidate aviation scene simulation image; Performing clustering comparison on the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed; If the similarity reaches or exceeds a similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

16. An electronic device, characterized in that: include: processor; as well as A memory is used to store a program of a method for processing an aerial scene image. After the device is powered on and the program of the method for processing an aerial scene image is run by the processor, the method according to any one of claims 1 to 14 is executed.

17. A storage medium, characterized in that: A program storing a method for processing an aerial scene image is executed by a processor to perform the method according to any one of claims 1 to 14.

18. A method for processing an aerial scene image, characterized in that: include: Obtaining a target real aerial scene image, and obtaining a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image; Obtaining annotation data of objects in the aerial scene simulation image to be processed; Obtaining a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image; Obtaining, according to the labeled data of the object, labeled data of the target object in the target aerial scene simulation image; The step of obtaining a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image comprises: Inputting the to-be-processed aviation scene simulation image and the target real aviation scene image into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image; The step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises: Inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real aerial scene image; Migrating the high-dimensional image feature data of the target real aviation scene image into the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image; The step of transferring the high-dimensional image feature data of the target real aviation scene image into the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image comprises: Obtaining initial high-dimensional image feature data of the to-be-processed aerial scene simulation image; According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image; The step of adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image comprises: According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain a candidate aviation scene simulation image; Performing clustering comparison on the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed; If the similarity reaches or exceeds a similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

19. The method for processing aerial scene images according to claim 18, characterized in that: The step of obtaining the labeling data of the target object in the target aerial scene simulation image according to the labeling data of the object comprises: Obtaining a matching relationship between the target object and the object; The labeled data of the target object is obtained according to the labeled data of the object and the matching relationship.

20. The method for processing aerial scene images according to claim 19, characterized in that: The annotation data of the object at least includes one or more of the position data of the object in the to-be-processed aerial scene simulation image, the name data of the object and the contour data of the object.

21. The method for processing aerial scene images according to claim 18, characterized in that: The method of obtaining a target aviation scene simulation image based on the aviation scene simulation image to be processed and the target real aviation scene image comprises: inputting the aviation scene simulation image to be processed and the target real aviation scene image into a target generative adversarial network model to obtain the target aviation scene simulation image.

22. The method for processing aerial scene images according to claim 21, characterized in that: The step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises: Obtaining a sample aerial scene simulation image and obtaining a sample real aerial scene image; The to-be-trained generative adversarial network model is trained according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model.

23. The method for processing aerial scene images according to claim 22, characterized in that: The method of performing model training on the to-be-trained generative adversarial network model according to the sample aviation scene simulation image and the sample real aviation scene image to obtain the target generative adversarial network model includes: Inputting a first sample aviation scene simulation image and a first sample real aviation scene image into the to-be-trained generative adversarial network model to obtain image feature data of the first sample real aviation scene image; Migrating the image feature data of the first sample real aviation scene image to the first sample aviation scene simulation image to obtain a first target aviation scene simulation image; Determining whether the image feature data of the first target aviation scene simulation image matches the image feature data of the first sample real aviation scene image; If so, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

24. The method for processing aerial scene images according to claim 23, characterized in that: Also includes: If the image feature data of the first target aviation scene simulation image does not match the image feature data of the first sample real aviation scene image, inputting the second sample aviation scene simulation image and the second sample real aviation scene image into the to-be-trained generative adversarial network model to obtain the image feature data of the second sample real aviation scene image; Migrating the image feature data of the second sample real aviation scene image to the second sample aviation scene simulation image to obtain a second target aviation scene simulation image; Determining whether the image feature data of the second target aviation scene simulation image matches the image feature data of the second sample real aviation scene image; If the image feature data of the second target aerial scene simulation image matches the image feature data of the second sample real aerial scene image, the to-be-trained generative adversarial network model is used as the target generative adversarial network model.

25. The method for processing aerial scene images according to claim 24, characterized in that: Also includes: If the image feature data of the second target aerial scene simulation image does not match the image feature data of the second sample real aerial scene image, the sample aerial scene simulation image and the sample real aerial scene image are sequentially input into the generative adversarial network model to be trained until the target generative adversarial network model is obtained.

26. The method for processing aerial scene images according to claim 18, characterized in that: The step of obtaining a target real aerial scene image and obtaining a simulated aerial scene image to be processed includes: obtaining a target object in the target real aerial scene image; Constructing a three-dimensional scene model corresponding to the target real aerial scene image according to the target real aerial scene image and the target object; Performing imaging simulation on the three-dimensional scene model corresponding to the aviation scene simulation image to be processed to obtain the aviation scene simulation image to be processed; The obtaining of the annotation data of the object in the to-be-processed aviation scene simulation image includes: obtaining the annotation data of the object according to a three-dimensional scene model corresponding to the to-be-processed aviation scene simulation image.

27. The method for processing aerial scene images according to claim 18, characterized in that: The simulated aviation scene image to be processed is a simulated image corresponding to the target real aviation scene image, including: the simulated aviation scene image to be processed is a simulated image generated by image simulation and corresponding to the target real aviation scene image.

28. A device for processing aerial scene images, characterized in that: include: An image acquisition unit, used to acquire a target real aerial scene image and to acquire a to-be-processed aerial scene simulation image, wherein the to-be-processed aerial scene simulation image is a simulation image corresponding to the target real aerial scene image; A labeling data obtaining unit, used for obtaining labeling data of objects in the aerial scene simulation image to be processed; A target image acquisition unit, configured to acquire a target aviation scene simulation image according to the to-be-processed aviation scene simulation image and the target real aviation scene image; A labeling data obtaining unit, used for obtaining labeling data of a target object in the target aerial scene simulation image according to the labeling data of the object; The target image obtaining unit is further used for: Inputting the to-be-processed aviation scene simulation image and the target real aviation scene image into a target generative adversarial network model to obtain a target aviation scene simulation image whose image feature data matches the image feature data of the target real aviation scene image; The step of inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain a target aerial scene simulation image whose image feature data matches the image feature data of the target real aerial scene image comprises: Inputting the to-be-processed aerial scene simulation image and the target real aerial scene image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real aerial scene image; Migrating the high-dimensional image feature data of the target real aviation scene image into the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image; The step of transferring the high-dimensional image feature data of the target real aviation scene image into the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image comprises: Obtaining initial high-dimensional image feature data of the to-be-processed aerial scene simulation image; According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image to obtain the target aviation scene simulation image; The step of adjusting the initial high-dimensional image feature data of the to-be-processed aviation scene simulation image according to the high-dimensional image feature data of the target real aviation scene image to obtain the target aviation scene simulation image comprises: According to the high-dimensional image feature data of the target real aviation scene image, adjusting the initial high-dimensional image feature data of the aviation scene simulation image to be processed to obtain a candidate aviation scene simulation image; Performing clustering comparison on the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate aviation scene simulation image and the initial high-dimensional image feature data of the aviation scene simulation image to be processed; If the similarity reaches or exceeds a similarity threshold, the candidate aviation scene simulation image is used as the target aviation scene simulation image.

29. An electronic device, characterized in that: include: processor; as well as The memory is used to store a program of a method for processing an aerial scene image. After the device is powered on and the program of the method for processing an aerial scene image is run by the processor, the method according to any one of claims 18 to 27 is executed.

30. A storage medium, characterized in that: A program storing a method for processing an aerial scene image is executed by a processor to perform the method according to any one of claims 18 to 27.

31. A method for processing an image, characterized in that: include: Obtaining a target real image and obtaining a simulated image to be processed, wherein the simulated image to be processed is a simulated image corresponding to the target real image; Inputting the simulated image to be processed and the target real image into a target generative adversarial network model to obtain a target simulated image whose image feature data matches the image feature data of the target real image; The step of inputting the simulated image to be processed and the target real image into a target generative adversarial network model to obtain a target simulated image whose image feature data matches the image feature data of the target real image comprises: Inputting the simulated image to be processed and the target real image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real image; Migrating the high-dimensional image feature data of the target real image into the simulated image to be processed to obtain the target simulated image; The step of transferring the high-dimensional image feature data of the target real image into the simulated image to be processed to obtain the target simulated image comprises: Obtaining initial high-dimensional image feature data of the simulated image to be processed; According to the high-dimensional image feature data of the target real image, adjusting the initial high-dimensional image feature data of the simulated image to be processed to obtain the target simulated image; The step of adjusting the initial high-dimensional image feature data of the simulated image to be processed according to the high-dimensional image feature data of the target real image to obtain the target simulated image comprises: According to the high-dimensional image feature data of the target real image, adjusting the initial high-dimensional image feature data of the simulated image to be processed to obtain a candidate simulated image; Performing clustering comparison on the high-dimensional image feature data of the candidate simulated image and the initial high-dimensional image feature data of the simulated image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate simulated image and the initial high-dimensional image feature data of the simulated image to be processed; If the similarity reaches or exceeds a similarity threshold, the candidate simulated image is used as the target simulated image.

32. A method for processing an image, characterized in that: include: Obtaining a target real image and obtaining a simulated image to be processed, wherein the simulated image to be processed is a simulated image corresponding to the target real image; Obtaining annotation data of objects in the simulated image to be processed; Obtaining a target simulated image according to the simulated image to be processed and the target real image; Obtaining, according to the labeled data of the object, labeled data of the target object in the target simulation image; The step of obtaining a target simulated image according to the simulated image to be processed and the target real image comprises: Inputting the simulated image to be processed and the target real image into a target generative adversarial network model to obtain a target simulated image whose image feature data matches the image feature data of the target real image; The step of inputting the simulated image to be processed and the target real image into a target generative adversarial network model to obtain a target simulated image whose image feature data matches the image feature data of the target real image comprises: Inputting the simulated image to be processed and the target real image into a target generative adversarial network model to obtain high-dimensional image feature data of the target real image; Migrating the high-dimensional image feature data of the target real image into the simulated image to be processed to obtain the target simulated image; The step of transferring the high-dimensional image feature data of the target real image into the simulated image to be processed to obtain the target simulated image comprises: Obtaining initial high-dimensional image feature data of the simulated image to be processed; According to the high-dimensional image feature data of the target real image, adjusting the initial high-dimensional image feature data of the simulated image to be processed to obtain the target simulated image; The step of adjusting the initial high-dimensional image feature data of the simulated image to be processed according to the high-dimensional image feature data of the target real image to obtain the target simulated image comprises: According to the high-dimensional image feature data of the target real image, adjusting the initial high-dimensional image feature data of the simulated image to be processed to obtain a candidate simulated image; Performing clustering comparison on the high-dimensional image feature data of the candidate simulated image and the initial high-dimensional image feature data of the simulated image to be processed, to obtain the similarity between the high-dimensional image feature data of the candidate simulated image and the initial high-dimensional image feature data of the simulated image to be processed; If the similarity reaches or exceeds a similarity threshold, the candidate simulated image is used as the target simulated image.

Images