Ground object recognition method, device, storage medium and terminal

Abstract

The embodiment of the present application discloses a method, device, storage medium and terminal for identifying land objects, which belongs to the field of computer technology. The method is applied to a terminal, and the terminal fuses a first remote sensing image and a second remote sensing image based on a principal component analysis (PCA) algorithm to obtain a fused image. The first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times. The first remote sensing image and the second remote sensing image contain different land objects. The fused image is processed based on a preset new land object recognition model to obtain the recognition result of the new land object, which can realize automatic recognition of the new land object, reduce labor costs, and effectively improve the efficiency of identifying the new land object.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, storage medium and terminal for identifying a ground object. Background Art

[0002] With the development of remote sensing technology, the use of remote sensing images for object recognition has gradually become an important application field in remote sensing technology, such as the recognition of houses, greenhouses, oil fields and other objects. In certain scenarios, it also involves the recognition of new objects, such as the recognition of new oil fields. In related technologies, the recognition of new objects in remote sensing images usually requires manual annotation of the remote sensing images pixel by pixel, which has high labor costs and a long processing cycle. In addition, remote sensing images are easily affected by weather, resulting in inaccurate image data, which in turn leads to the inability to accurately annotate new objects. Summary of the invention

[0003] The embodiments of the present application provide a method, device, storage medium and terminal for identifying objects, which can solve the problem of high labor cost for marking new objects in related technologies. The technical solution is as follows:

[0004] In a first aspect, an embodiment of the present application provides a method for identifying a ground object, the method comprising:

[0005] The first remote sensing image and the second remote sensing image are fused based on a principal component analysis (PCA) algorithm to obtain a fused image; wherein the first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times;

[0006] The fused image is processed based on a preset new object recognition model to obtain a recognition result of the new object.

[0007] In a second aspect, an embodiment of the present application provides a ground object recognition device, the device comprising:

[0008] A first processing module is used to fuse the first remote sensing image and the second remote sensing image based on a principal component analysis (PCA) algorithm to obtain a fused image; wherein the first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times;

[0009] The second processing module is used to process the fused image based on a preset new feature recognition model to obtain a recognition result of the new feature.

[0010] In a third aspect, an embodiment of the present application provides a computer storage medium, wherein the computer storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the above-mentioned method steps.

[0011] In a fourth aspect, an embodiment of the present application provides a terminal, comprising: a processor, a memory, and a display screen; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the above-mentioned method steps.

[0012] The beneficial effects brought about by the technical solutions provided by some embodiments of the present application include at least:

[0013] When the scheme of the embodiment of the present application is executed, the terminal fuses the first remote sensing image and the second remote sensing image based on the principal component analysis (PCA) algorithm to obtain a fused image. The first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times. The first remote sensing image and the second remote sensing image contain different ground objects. The fused image is processed based on a preset new ground object recognition model to obtain an recognition result of the new ground object, which can realize automatic recognition of the new ground objects, reduce labor costs, and effectively improve the efficiency of identifying the new ground objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 It is a schematic diagram of the system architecture of a method for identifying a ground object provided in an embodiment of the present application;

[0016] Figure 2 It is a flowchart of a method for identifying a ground object provided in an embodiment of the present application;

[0017] Figure 3 is another flowchart of a method for identifying a ground object provided in an embodiment of the present application;

[0018] Figure 4 It is a schematic diagram of a Cascade R-CNN model provided in an embodiment of the present application;

[0019] Figure 5 is another flowchart of a method for identifying a ground object provided in an embodiment of the present application;

[0020] Figure 6 This is a schematic diagram of the newly added ground feature recognition effect provided in the embodiment of the present application;

[0021] Figure 7 It is a structural schematic diagram of a ground object recognition device provided in an embodiment of the present application;

[0022] Figure 8It is a structural diagram of a terminal involved in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to make the objectives, technical solutions and advantages of the present application more clear, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0024] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application.

[0025] In the description of this application, it should be understood that the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previously associated objects are in an "or" relationship.

[0026] In order to make the objectives, technical solutions and advantages of the present application more clear, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0027] See also Figure 1 , which shows a schematic diagram of a system architecture for ground object identification provided in an embodiment of the present application, including a remote sensing satellite 101, a remote sensing satellite ground station 102 and a terminal device 103, as well as a target object 104 of which the remote sensing satellite 101 takes a remote sensing image. The remote sensing satellite 101 will take a remote sensing image of the target object 104 and transmit the taken remote sensing image to the terminal device 103 through the remote sensing satellite ground station 102.

[0028] Usually, the staff of the remote sensing satellite ground station 102 will complete the allocation of all resources according to the task requirements, calculate the satellite orbits, determine the time when different satellites pass through the five remote sensing satellite ground stations 102, and arrange the antenna, recording and transmission resources of the remote sensing satellite ground station 102 according to the scientific tasks. When the remote sensing satellite 101 passes over the remote sensing satellite ground station 102, the remote sensing satellite ground station 102 will convert the collected data into electromagnetic waves suitable for propagation in free space and send them to the remote sensing satellite ground station. The remote sensing satellite ground station 102 will align the antenna with the direction where the remote sensing satellite 101 is about to appear. When the remote sensing satellite 101 appears and starts to send electromagnetic wave signals, the corresponding remote sensing satellite ground station antenna will rotate continuously to achieve full locking and tracking of electromagnetic waves. At the same time, the remote sensing satellite ground station 102 amplifies, converts the frequency, demodulates and other processes on the received electromagnetic wave signals, and sends the output satellite raw data baseband signal to the terminal device 103, and the terminal device 103 will further extract and analyze the received data.

[0029] Remote sensing satellite 101 is an artificial satellite used as a remote sensing platform in outer space. The remote sensing technology using satellite as a platform is called satellite remote sensing. Usually, remote sensing satellite 101 can operate in orbit for several years. The satellite orbit can be determined according to needs. Remote sensing satellite 101 can cover the entire earth or any designated area within a specified time. When operating along the geosynchronous orbit, it can continuously perform remote sensing of a designated area on the earth's surface. All remote sensing satellites need remote sensing satellite ground stations. The satellite data obtained from the remote sensing market platform can monitor agriculture, forestry, ocean, land, environmental protection, meteorology and other conditions. Remote sensing satellites are mainly of three types: meteorological satellites, land satellites and ocean satellites.

[0030] The remote sensing satellite ground station 102 is also called a satellite ground receiving station. Its main task is to capture and track satellites, receive, demodulate and record satellite remote sensing data and auxiliary data, and monitor and judge the working conditions of satellite remote sensors and their transmission systems in real time through a quick-view system, and evaluate image quality. Within the antenna tracking and receiving range, the remote sensing satellite ground receiving station can directly receive the remote sensing data that the satellite is collecting and sending in real time. The remote sensing satellite ground station 102 is mainly composed of a parabolic antenna, a feed source, a high-frequency head, and a satellite receiving card (satellite receiver). The parabolic antenna reflects the energy of satellite signals from the air and converges them into a point (focus); the feed source is a speaker for collecting satellite signals at the focus of the parabolic antenna. It is the source of feed energy and can collect all the energy gathered at the focus. The high-frequency head (LNB) down-converts and amplifies the satellite signal sent by the feed source, and then transmits it to the satellite receiver; the satellite receiving card (satellite receiver) demodulates the satellite signal transmitted by the high-frequency head and demodulates the data signal or satellite TV image signal and accompanying sound signal.

[0031] The terminal device 103 may be any electronic device with a display screen, including but not limited to a tablet computer, a portable computer, and a desktop computer, etc. The terminal device 103 has a preset newly added object recognition model built in. The terminal device 103 may recognize newly added objects in the acquired remote sensing images, and realize the recognition of newly added objects in the remote sensing images of the same geographical area at multiple different times. The terminal device 103 may be hardware or software. When the terminal device 103 is software, it may be installed in the electronic devices listed above, and it may be implemented as multiple software or software modules, or it may be implemented as a single software or software module, which is not specifically limited here. When the terminal device 103 is hardware, a display device may also be installed thereon, and the display device display may be any device capable of realizing a display function, such as: the display device may be a cathode ray tube display (Cathode ray tube display, referred to as CR), a light-emitting diode display (Light-emitting diode display, referred to as LED), an electronic ink screen, a liquid crystal display (Liquid crystal display, referred to as LCD), a plasma display panel (Plasma display panel, referred to as PDP), etc.

[0032] It should be understood that Figure 1 The number of remote sensing satellites, remote sensing satellite ground stations and terminal devices in the figure is only for illustration. According to actual needs, any number of remote sensing satellites, remote sensing satellite ground stations and terminal devices can be used.

[0033] In the following method embodiments, for the sake of convenience, only the execution subject of each step is introduced and described as the terminal.

[0034] The following will be combined with the attached Figures 2 to 5 , the ground object recognition method provided in the embodiment of the present application is introduced in detail.

[0035] See also Figure 2 , which is a flow chart of a method for identifying a ground object according to an embodiment of the present application. This embodiment is illustrated by applying a method for identifying a ground object to a terminal. The method for identifying a ground object may include the following steps:

[0036] S201, fusing the first remote sensing image and the second remote sensing image based on a principal component analysis (PCA) algorithm to obtain a fused image.

[0037] Among them, remote sensing images refer to films or photos that record the size of electromagnetic waves of various objects, mainly divided into aerial photographs and satellite photos. Remote sensing images processed by computers are digital images obtained after analog / digital (A / D) conversion; the first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times. The first remote sensing image and the second remote sensing image can be two remote sensing images containing the same objects, or two remote sensing images containing different objects (such as: the second remote sensing image contains new objects compared to the first remote sensing image). Fusion image refers to the image obtained by processing the first remote sensing image and the second remote sensing image with the PCA algorithm, which contains the information of the first remote sensing image and the second remote sensing image. Principal Component Analysis (PCA) is a statistical method that uses mathematical dimensionality reduction to convert a set of variables that may be correlated into a set of linearly uncorrelated variables through orthogonal transformation, and the converted variables can represent the information of the original variables. The converted set of variables is called principal components.

[0038] Generally, the terminal obtains two remote sensing images of the same geographical area at different times that need to identify new features, namely, the first remote sensing image and the second remote sensing image. The first remote sensing image can be cropped to obtain multiple first small-block images, and the second remote sensing image can be cropped to obtain multiple second small-block images. The multiple first small-block images and the multiple second small-block images are preprocessed separately; wherein the first small-block images correspond one-to-one to the second small-block images, and the preprocessing includes at least one of the following: stretching, brightness adjustment, and hue adjustment. Based on the PCA algorithm, the multiple first small-block images and the multiple second small-block images are fused to obtain multiple fused images, the number of the first small-block images is consistent with the number of the second small-block images, and the number of fused images is also consistent with the number of the first small-block images and the second small-block images.

[0039] Supplementary explanation: the terminal includes but is not limited to processing two remote sensing images of the same geographical area at different times, and can also process multiple remote sensing images of the same geographical area at different times. When processing remote sensing images at different times, any two remote sensing images of the same geographical area at different times are identified. For example, the terminal can process remote sensing images at three different times t1, t2, and t3, and the remote sensing image at time t2 contains new features compared to the remote sensing image at time t1, and the remote sensing image at time t3 contains new features relative to the remote sensing image at time t1. Then the terminal can output the new feature location information of the remote sensing image at time t2 relative to time t1, the new feature location information of the remote sensing image at time t3 relative to time t1, and the new feature location information of the remote sensing image at time t3 relative to time t2.

[0040] The image will retain the information of the original data to the maximum extent while reducing the data dimension through the PCA algorithm. The main idea of ​​fusing the first remote sensing image and the second remote sensing image based on the PCA algorithm is: first, perform principal component transformation on the multispectral image, and then use the stretched high spatial resolution image to replace the first principal component for inverse principal component transformation to obtain a fused image. Using the PCA algorithm to process the first remote sensing image and the second remote sensing image can obtain a fused image with better image spectral characteristics, thus achieving image enhancement.

[0041] S202: Process the fused image based on a preset new object recognition model to obtain a recognition result of the new object.

[0042] The recognition result refers to whether the fused image contains newly added location information.

[0043] Generally, the preset new feature recognition model can automatically recognize the new features in the fused image and obtain the recognition result of the new features. The first remote sensing image and the second remote sensing image can be two remote sensing images containing the same features, or two remote sensing images containing different features (such as: the second remote sensing image contains new features compared to the first remote sensing image). When the first remote sensing image and the second remote sensing image contain different features, the recognition result of the new features is the location information of the new features in the second remote sensing image compared to the first remote sensing image, and the location information refers to the coordinate information corresponding to the new features. When the first remote sensing image and the second remote sensing image contain the same features, the recognition result of the new features is the location information without the new features, and a reminder message is displayed through the display unit. The reminder message is used to remind that there are no new features in the second remote sensing image compared to the first remote sensing image.

[0044] The preset new object recognition model is obtained by training sample images at multiple times. The training process of the preset new object recognition model may include:

[0045] The first sample image and the second sample image are obtained. The first sample image and the second sample image are remote sensing images of the same geographical area at different times. The first sample image and the second sample image contain different objects. The first sample image is cropped to obtain multiple first sample small-block images, and the second sample image is cropped to obtain multiple second sample small-block images. The first sample small-block images and the second sample small-block images are in one-to-one correspondence. The multiple first sample small-block images and the multiple second sample small-block images are preprocessed respectively, and the multiple preprocessed first sample small-block images and the multiple preprocessed second sample small-block images are fused based on the PCA algorithm to obtain multiple sample fused images. The preprocessing includes at least one of the following: stretching, brightness adjustment, and hue adjustment. Based on a preset object recognition model, multiple fused and annotated images are trained to obtain a preset newly added object recognition model. The preset object recognition model is obtained by training the sample image at a single moment. The fused and annotated image is obtained by manually annotating the sample fused image. The manual annotation process refers to the position annotation of the newly added objects in the sample fused image.

[0046] The training process of the preset object recognition model may include:

[0047] A third sample image is obtained, and the third sample image is cropped to obtain a plurality of third sample small-block images. The plurality of third sample small-block images are preprocessed, and the preprocessing includes at least one of the following: stretching, brightness adjustment, and hue adjustment. A preset ground object recognition model is obtained by training a plurality of sample annotated images based on the target detection Cascade R-CNN model, and the sample annotated images are obtained by manually annotating the preprocessed third sample small-block images, and the sample annotated images contain the location identifiers of the ground objects.

[0048] In another possible implementation, the training process of the preset new ground object recognition model may include:

[0049] Based on the PCA algorithm, the first sample image and the second sample image are fused to obtain multiple sample fused images. The first sample image and the second sample image are remote sensing images of the same geographical area at different times. The first sample image and the second sample image contain different objects. The multiple sample fused images are preprocessed, and the multiple fused annotated images are trained based on the target detection Cascade R-CNN model to obtain the preset new object recognition model. The fused annotated images are obtained by manually annotating the sample fused images. The manual annotation process refers to the location annotation of the new objects in the sample fused images.

[0050] From the above content, it can be seen that in the method for identifying objects provided by this scheme, the terminal fuses the first remote sensing image and the second remote sensing image based on the principal component analysis PCA algorithm to obtain a fused image, the first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times, the first remote sensing image and the second remote sensing image contain different objects, and the fused image is processed based on a preset new object recognition model to obtain an identification result of the new object, the first remote sensing image and the second remote sensing image contain different objects, and the identification result of the new object is the position information of the new object in the second remote sensing image compared with the first remote sensing image, which can realize automatic identification of the new objects, reduce labor costs, and effectively improve the efficiency of identifying the new objects.

[0051] See also Figure 3 , which is another flow chart of a method for identifying a ground object provided in an embodiment of the present application. This embodiment is illustrated by applying a method for identifying a ground object in a terminal. The method for identifying a ground object may include the following steps:

[0052] S301, acquiring a first remote sensing image and a second remote sensing image.

[0053] Among them, remote sensing images refer to films or photos that record the size of electromagnetic waves of various objects, which are mainly divided into aerial photographs and satellite photos. Remote sensing images processed by computers are digital images obtained after analog / digital (A / D) conversion; the first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times. The first remote sensing image and the second remote sensing image can be two remote sensing images containing the same objects, or two remote sensing images containing different objects (such as: the second remote sensing image contains new objects compared to the first remote sensing image). Fusion image refers to the image obtained by processing the first remote sensing image and the second remote sensing image with the PCA algorithm, which contains the information of the first remote sensing image and the second remote sensing image. Usually, the first remote sensing image and the second remote sensing image have a certain span in time, so that new objects may appear in the area where the first remote sensing image and the second remote sensing image are located.

[0054] S302: cropping the first remote sensing image to obtain a plurality of first small images, and cropping the second remote sensing image to obtain a plurality of second small images.

[0055] The first small image is a small image obtained by randomly cutting the first remote sensing image, and the first remote sensing image can be cut randomly to obtain a plurality of small images corresponding to the first remote sensing image. The second small image is a small image obtained by randomly cutting the second remote sensing image, and the second remote sensing image can be cut randomly to obtain a plurality of small images corresponding to the second remote sensing image. The geographical location contained in the first small image corresponds to the geographical location contained in the second small image.

[0056] Generally, the original remote sensing image can be cropped to obtain multiple small images corresponding to the original remote sensing image. The process of cropping the remote sensing image to obtain small images is the process of intercepting a part of the original remote sensing image to generate a new image.

[0057] S303, pre-processing the plurality of first small-block images and the plurality of second small-block images respectively.

[0058] The preprocessing includes at least one of the following: stretching, brightness adjustment, and hue adjustment.

[0059] Generally, the remote sensing images collected at different times are affected by the seasons, weather, cloud cover, orthorectification and other changes during the process of collecting remote sensing images, resulting in inconsistency in brightness, hue, angle, etc. The cropped multiple first small-block images and multiple second small-block images can be subjected to pre-processing operations such as stretching, brightness adjustment, and hue adjustment, so that the brightness, hue, and image angle of the multiple first small-block images and multiple second small-block images can be unified, which is convenient for subsequent fusion processing of the multiple first small-block images and multiple second small-block images.

[0060] S304, fusing the plurality of first small-block images and the plurality of second small-block images based on a PCA algorithm to obtain a fused image.

[0061] The fused image refers to an image obtained by processing the first remote sensing image and the second remote sensing image with the PCA algorithm, that is, an image obtained by processing multiple first small-block images and multiple second small-block images with the PCA algorithm, which contains the information of the first remote sensing image and the second remote sensing image. The number of first small-block images is the same as the number of second small-block images, and the number of fused images is also the same as the number of first small-block images and the number of second small-block images.

[0062] Generally, based on the PCA algorithm, image fusion processing can be performed on multiple first small-block images and multiple second small-block images, which can ensure that the brightness and color tone of multiple fused images obtained by fusing multiple first small-block images with multiple second small-block images are consistent, and can significantly highlight the newly added objects in the fused images.

[0063] When the image is processed by the PCA algorithm to reduce the data dimension, the information of the original data will be retained to the maximum extent. The main idea of ​​fusing multiple first small images and multiple second small images based on the PCA algorithm is: first, the multispectral image is transformed into principal components, and then the stretched high spatial resolution image is used to replace the first principal component for inverse principal component transformation to obtain a fused image. The PCA algorithm is used to process multiple first small images and multiple second small images to obtain a fused image with good image spectral characteristics, thereby achieving image enhancement. Usually, multiple first small images and multiple second small images can be 512×512 pixel, 4-channel images. After the PCA algorithm is used for dimensionality reduction, the multiple fused images obtained are 512×512 pixel, 3-channel images; 4 channels refer to four bands: blue band, green band, red band and infrared band, and 3 channels refer to three bands: blue band, green band and red band.

[0064] S305 , obtaining a third sample image, and performing cropping processing on the third sample image to obtain a plurality of third sample small block images.

[0065] The third sample image refers to a remote sensing image containing a large number of objects at a single moment, which is used as sample data for training a preset object recognition model at a single moment. The third sample small block image refers to a small block image obtained by randomly cutting the third sample image.

[0066] S306, pre-processing the plurality of third sample small-block images, and training the plurality of sample labeled images based on the target detection Cascade R-CNN model to obtain a preset ground object recognition model.

[0067] The preprocessing includes at least one of the following: stretching, brightness adjustment, and hue adjustment. The sample annotated image is obtained by manually annotating the preprocessed third sample small block image, and the sample annotated image contains the location mark of the ground object.

[0068] Generally, multiple third sample small-block images are preprocessed so that the multiple third sample small-block images can unify brightness and tone, and the sizes of the multiple third sample small-block images after preprocessing are multiple RGB small-block images of 512×512 pixels and 3 channels. The multiple third sample small-block images after preprocessing are annotated with objects by manually annotating objects, thereby obtaining sample data that can be used to train a preset object recognition model at a single moment, that is, a sample annotated image. Using the sample annotated images as sample data, iterative training of the Cascade R-CNN model can generate an object recognition model that can recognize objects, that is, a preset object recognition model. The Cascade R-CNN model used in this application can be found in Figure 4 .

[0069] The Cascade R-CNN model is a cascade detection model composed of a series of detection models, and each detection model is trained based on positive and negative samples with different Intersection Over Union (IOU) thresholds. The output of the previous detection model is used as the input of the next detection model, and the threshold of the detection model at the later end of the cascade is larger. The Cascade R-CNN model performs better than detection models trained based on other thresholds, so the IOU threshold used by each detection model should be as close as possible to the IOU threshold of the input candidate box. The Cascade R-CNN model adopts a cascade approach that allows the detection model at each stage to focus on detecting candidate boxes with an IOU threshold within a certain range. Because the output IOU threshold is generally larger than the input IOU threshold, the detection effect will get better and better.

[0070] S307, acquiring a first sample image and a second sample image.

[0071] Among them, the first sample image and the second sample image are sample remote sensing images of the same geographical area at different times. There is a certain span between the time point corresponding to the first sample image and the time point corresponding to the second sample image, so that new land features will appear in the geographical area within the time period (the time period from the time point corresponding to the first sample image to the time point corresponding to the second sample image).

[0072] S308 , cutting the first sample image to obtain a plurality of first sample small-block images, and cutting the second sample image to obtain a plurality of second sample small-block images.

[0073] The first sample small block image is a sample small block image obtained by randomly cutting the first sample image, and the first sample image can be cut randomly to obtain a plurality of sample small block images corresponding to the first sample image. The second sample small block image is a sample small block image obtained by randomly cutting the second sample image, and the second sample image can be cut randomly to obtain a plurality of sample small block images corresponding to the second sample image.

[0074] Generally, multiple original sample remote sensing images at different times can be cropped to obtain multiple sample small images corresponding to the original sample remote sensing images. The process of cropping sample remote sensing images to obtain sample small images is also the process of intercepting a part of the original sample remote sensing image to generate a new sample image.

[0075] S309, preprocessing the plurality of first sample small block images and the plurality of second sample small block images respectively, and fusing the preprocessed plurality of first sample small block images and the preprocessed plurality of second sample small block images based on the PCA algorithm to obtain a plurality of sample fused images.

[0076] The sample fusion image refers to an image obtained by processing a plurality of first sample small-block images and a plurality of second sample small-block images by the PCA algorithm, and contains the information of the first sample images and the second sample images. The number of the first sample small-block images is consistent with the number of the second sample small-block images, and the number of the sample fusion images is also consistent with the number of the first sample small-block images and the number of the second sample small-block images.

[0077] Generally, the cropped multiple first sample small block images and multiple second sample small block images are subjected to preprocessing operations such as stretching, brightness adjustment, and hue adjustment, so that the brightness, hue, and image angle of the multiple first sample small block images and the multiple second sample small block images can be unified, which is convenient for subsequent fusion processing of the multiple first sample small block images and the multiple second sample small block images. Based on the PCA algorithm, the multiple first sample small block images and the multiple second sample small block images can be subjected to image fusion processing, which can ensure that the brightness and hue of the multiple fused images obtained by fusion of the multiple first sample small block images and the multiple second sample small block images are consistent, and can significantly highlight the newly added objects in the fused image.

[0078] S310, training multiple fused and annotated images based on a preset land object recognition model to obtain a preset new land object recognition model.

[0079] Among them, the preset feature recognition model is trained by the sample image (the third sample image) at a single moment, and the fused labeled image is obtained after the sample fused image is manually labeled. The manual labeling process refers to the position labeling of the newly added features in the sample fused image.

[0080] Generally, the plurality of first sample small block images and the plurality of second sample small block images are preprocessed respectively so that the plurality of first sample small block images and the plurality of second sample small block images can unify the brightness and hue. The plurality of first sample small block images and the plurality of second sample small block images after preprocessing are all 512×512 pixels and 4-channel plurality of sample small block images. After the plurality of first sample small block images and the plurality of second sample small block images are subjected to PCA fusion processing, a plurality of sample fusion images can be obtained. Then, the plurality of first sample small block images and the plurality of second sample small block images after preprocessing can be respectively annotated with newly added objects by manually annotating newly added objects, thereby obtaining sample data that can be used to train the preset newly added object recognition model at multiple moments, that is, the fusion annotation image. Using the fusion annotation image as sample data, iterative training of the pre-trained preset object recognition model can generate a newly added object recognition model that can recognize newly added objects, that is, the preset newly added object recognition model. The preset new feature recognition module can also perform new feature recognition on multiple remote sensing images of the same geographical area at different times, and is not limited to the above-mentioned new feature recognition on remote sensing images of the same geographical area at two times.

[0081] S311, processing the fused image based on a preset new object recognition model to obtain a recognition result of the new object.

[0082] The recognition result refers to whether the fused image contains newly added location information.

[0083] Generally, new objects are identified based on a pre-trained new object identification model for multiple fused images after PCA fusion processing. If the fused image contains new objects, the identification result of the new objects can be obtained as the location information of the new objects in the second remote sensing image compared with the first remote sensing image; if the fused image does not contain new objects, the identification result of the new objects is the location information of the new objects, and a reminder message is displayed through the display unit, which is a reminder message about the fact that there are no new objects in the second remote sensing image compared with the first remote sensing image.

[0084] For example: the remote sensing image at time t1 and the remote sensing image at time t2 are two remote sensing images of the same geographical area at different times. The terminal processes the remote sensing image at time t1 and the remote sensing image at time t2 based on the preset new object recognition model. If the remote sensing image at time t2 has a new object compared to the remote sensing image at time t1, the location information of the new object will be obtained; if the remote sensing image at time t2 has no new object compared to the remote sensing image at time t1, the reminder message "No new object appears at present" will be displayed through the display unit of the terminal.

[0085] In the training strategy of the model, the embodiment of the present application uses transfer learning to optimize the preset object recognition model at a single time point, so as to quickly train the preset new object recognition model at multiple time points, which can greatly reduce the number of samples required to be labeled and reduce labor costs. At the same time, based on the labeled samples of the preset object recognition model at a single time point, the labeled samples of the multi-time point model can be quickly generated, which can effectively reduce the time cost of training the model and labeling the samples.

[0086] From the above content, it can be seen that in the method for identifying objects provided by this solution, the terminal obtains a first remote sensing image and a second remote sensing image, crops the first remote sensing image to obtain a plurality of first small-block images, crops the second remote sensing image to obtain a plurality of second small-block images, pre-processes the plurality of first small-block images and the plurality of second small-block images respectively, fuses the plurality of first small-block images and the plurality of second small-block images based on the PCA algorithm to obtain a fused image, obtains a third sample image, crops the third sample image to obtain a plurality of third sample small-block images, pre-processes the plurality of third sample small-block images, and performs target detection Cascade based on the target detection Cascade. The R-CNN model trains multiple sample annotated images to obtain a preset object recognition model, obtains a first sample image and a second sample image, cuts the first sample image to obtain multiple first sample small-block images, and cuts the second sample image to obtain multiple second sample small-block images, respectively preprocesses the multiple first sample small-block images and the multiple second sample small-block images, and fuses the preprocessed multiple first sample small-block images and the preprocessed multiple second sample small-block images based on the PCA algorithm to obtain multiple sample fused images, trains multiple fused annotated images based on the preset object recognition model to obtain a preset new object recognition model, processes the fused images based on the preset new object recognition model to obtain the recognition result of the new object, and the objects contained in the first remote sensing image and the second remote sensing image are different, and the recognition result of the new object is the location information of the new object in the second remote sensing image compared with the first remote sensing image. In this way, the new objects in the remote sensing image can be automatically recognized, labor costs and time can be saved, and the accuracy and efficiency of recognizing the new objects can be improved.

[0087] See also Figure 5 , which is another flow chart of a method for identifying a ground object provided in an embodiment of the present application. This embodiment is illustrated by applying a method for identifying a ground object in a terminal. The method for identifying a ground object may include the following steps:

[0088] S501, acquiring a first remote sensing image and a second remote sensing image.

[0089] Please refer to the above step S301, which will not be repeated here.

[0090] S502: Crop the first remote sensing image to obtain a plurality of first small images, and crop the second remote sensing image to obtain a plurality of second small images.

[0091] Please refer to the above step S302, which will not be repeated here.

[0092] S503, pre-processing the plurality of first small-block images and the plurality of second small-block images respectively.

[0093] Please refer to the above step S303, which will not be repeated here.

[0094] S504, fusing the plurality of first small-block images and the plurality of second small-block images based on a PCA algorithm to obtain a fused image.

[0095] Please refer to the above step S304, which will not be repeated here.

[0096] S505 , performing fusion processing on the first sample image and the second sample image based on the PCA algorithm to obtain a plurality of sample fusion images.

[0097] Among them, the first sample image and the second sample image are sample remote sensing images of the same geographical area at different times. There is a certain span between the time point corresponding to the first sample image and the time point corresponding to the second sample image, so that within this time period (the time period from the time point corresponding to the first sample image to the time point corresponding to the second sample image), there will be new features appearing in this geographical area. The sample fusion image refers to an image obtained by processing multiple first sample small-block images and multiple second sample small-block images with the PCA algorithm, and contains the information of the first sample image and the second sample image. The number of first sample small-block images is consistent with the number of second sample small-block images, and the number of sample fusion images is also consistent with the number of first sample small-block images and the number of second sample small-block images.

[0098] Generally, after the terminal obtains the first sample image and the second sample image, it performs cropping processing on the first sample image to obtain multiple first sample small-block images, and cropping processing on the second sample image to obtain multiple second sample small-block images, that is, it performs cropping processing on multiple original sample remote sensing images at different times to obtain multiple sample small-block images corresponding to the original sample remote sensing images. The process of cropping the sample remote sensing images to obtain sample small-block images is the process of intercepting a certain part of the original sample remote sensing image to generate a new sample image. Based on the PCA algorithm, multiple first sample small-block images and multiple second sample small-block images can be fused to ensure that the brightness and color tone of the multiple fused images obtained by fusion of multiple first sample small-block images and multiple second sample small-block images are consistent, and the newly added objects in the fused images can be significantly highlighted.

[0099] S506, pre-processing the multiple sample fused images, and training the multiple fused labeled images based on the target detection Cascade R-CNN model to obtain a preset new object recognition model.

[0100] The fused labeled image is obtained by manually labeling the sample fused image, and the manual labeling process refers to labeling the positions of the newly added features in the sample fused image.

[0101] Generally, multiple sample fused images can be obtained after PCA fusion processing is performed on multiple first sample small-block images and multiple second sample small-block images. The multiple sample fused images are preprocessed by stretching, brightness adjustment, hue adjustment, etc., so that the brightness, hue and image angle of the multiple sample fused images can be unified. The multiple sample fused images that have been preprocessed are annotated with objects by manually annotating objects, thereby obtaining sample data that can be used to train a preset new object recognition model at multiple moments, that is, a fused annotated image. Using the fused annotated image as sample data, iterative training of the CascadeR-CNN model can generate an object recognition model that can recognize new objects, that is, a preset new object recognition model.

[0102] The Cascade R-CNN model is a cascade detection model composed of a series of detection models, and each detection model is trained based on positive and negative samples with different Intersection Over Union (IOU) thresholds. The output of the previous detection model is used as the input of the next detection model, and the threshold of the detection model at the later end of the cascade is larger. The Cascade R-CNN model performs better than detection models trained based on other thresholds, so the IOU threshold used by each detection model should be as close as possible to the IOU threshold of the input candidate box. The Cascade R-CNN model adopts a cascade approach that allows the detection model at each stage to focus on detecting candidate boxes with an IOU threshold within a certain range. Because the output IOU threshold is generally larger than the input IOU threshold, the detection effect will get better and better.

[0103] S507: Process the fused image based on a preset new object recognition model to obtain a recognition result of the new object.

[0104] Please refer to the above step S311, which will not be repeated here.

[0105] From the above content, it can be seen that in the method for identifying objects provided by this solution, the terminal obtains a first remote sensing image and a second remote sensing image, crops the first remote sensing image to obtain multiple first small images, and crops the second remote sensing image to obtain multiple second small images, pre-processes the multiple first small images and the multiple second small images respectively, fuses the multiple first small images and the multiple second small images based on the PCA algorithm to obtain a fused image, fuses the first sample image and the second sample image based on the PCA algorithm to obtain multiple sample fused images, pre-processes the multiple sample fused images, and trains the multiple fused and labeled images based on the target detection Cascade R-CNN model to obtain a preset new object recognition model, processes the fused image based on the preset new object recognition model to obtain the recognition result of the new object, the objects contained in the first remote sensing image and the second remote sensing image are different, and the recognition result of the new object is the position information of the new object in the second remote sensing image compared with the first remote sensing image, thereby realizing automatic recognition of the new objects in the remote sensing image, reducing labor costs, and improving the accuracy and efficiency of identifying the new objects.

[0106] See also Figure 6 The schematic diagram of the newly added object recognition effect shown in the figure, the left figure and the right figure are respectively derived from remote sensing images of the same geographical area at different times (the left figure corresponds to the first remote sensing image, and the right figure corresponds to the second remote sensing image), wherein the time point of the second remote sensing image is located after the first remote sensing image, and newly added objects appear in the second remote sensing image. The left figure is an effect diagram of the pre-trained object recognition model (preset object recognition model) being able to recognize the objects in the remote sensing image. The preset object recognition model can recognize all objects in the first remote sensing image, such as the object 601 marked with a box in the left figure, and other marked objects can also be seen in the left figure. The right picture is an effect picture of the pre-trained new feature recognition model (the preset new feature recognition model) being able to recognize the new features in the remote sensing image. The preset new feature recognition model can recognize the new features in the second remote sensing image compared to the features in the first remote sensing image, such as the new feature 603 marked with a box in the right picture. The original feature 602 in the left picture, as well as other marked new features and other original features can also be seen in the right picture.

[0107] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0108] See also Figure 7, which shows a schematic diagram of the structure of a ground object recognition device provided by an exemplary embodiment of the present application, hereinafter referred to as recognition device 7. Recognition device 7 can be implemented as all or part of the terminal through software, hardware or a combination of both. Recognition device 7 is applied to the terminal, and recognition device 7 includes:

[0109] The first processing module 701 is used to fuse the first remote sensing image and the second remote sensing image based on the principal component analysis (PCA) algorithm to obtain a fused image; wherein the first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times;

[0110] The second processing module 702 is used to process the fused image based on a preset new object recognition model to obtain a recognition result of the new object.

[0111] Optionally, the first remote sensing image and the second remote sensing image of the recognition device 7 contain different ground objects, and the recognition result of the newly added ground object is the position information of the newly added ground object in the second remote sensing image compared with the first remote sensing image.

[0112] Optionally, the first remote sensing image and the second remote sensing image of the recognition device 7 contain the same ground object, and the reminder information is displayed through a display unit.

[0113] Optionally, the first processing module 701 further includes:

[0114] A first acquisition unit, used for acquiring the first remote sensing image and the second remote sensing image;

[0115] A first processing unit is used to perform cropping processing on the first remote sensing image to obtain a plurality of first small-block images, and to perform cropping processing on the second remote sensing image to obtain a plurality of second small-block images;

[0116] The second processing unit is used to perform fusion processing on the multiple first small-block images and the multiple second small-block images based on the PCA algorithm to obtain the fused image.

[0117] Optionally, the first processing module 701 further includes:

[0118] A preprocessing unit is used to preprocess the multiple first small-block images and the multiple second small-block images respectively; wherein the first small-block images correspond to the second small-block images one by one, and the preprocessing includes at least one of the following: stretching, brightness adjustment, and hue adjustment.

[0119] Optionally, the identification device 7 further includes:

[0120] A third processing unit is used to perform fusion processing on the first sample image and the second sample image based on the PCA algorithm to obtain a plurality of sample fusion images; the first sample image and the second sample image are remote sensing images of the same geographical area at different times, and the first sample image and the second sample image contain different ground objects;

[0121] The first training unit is used to train multiple fused and annotated images based on a preset object recognition model to obtain the preset new object recognition model; wherein the preset object recognition model is obtained by training the sample image at a single moment, and the fused and annotated image is obtained by manually annotating the sample fused image, and the manual annotation processing refers to the position annotation of the new objects in the sample fused image.

[0122] Optionally, the identification device 7 further includes:

[0123] acquiring the first sample image and the second sample image;

[0124] A second acquisition unit, configured to perform cropping processing on the first sample image to obtain a plurality of first sample small-block images, and to perform cropping processing on the second sample image to obtain a plurality of second sample small-block images;

[0125] The fourth processing unit is used to preprocess the multiple first sample small block images and the multiple second sample small block images respectively, and fuse the preprocessed multiple first sample small block images and the preprocessed multiple second sample small block images based on the PCA algorithm to obtain the multiple sample fused images.

[0126] Optionally, the identification device 7 further includes:

[0127] A third acquisition unit, used for acquiring a third sample image, and performing cropping processing on the third sample image to obtain a plurality of third sample small block images;

[0128] The fifth processing unit is used to preprocess the multiple third sample small-block images, and train the multiple sample annotated images based on the target detection Cascade R-CNN model to obtain the preset ground object recognition model; wherein the sample annotated images are obtained by manually annotating the preprocessed third sample small-block images, and the sample annotated images contain the location identification of the ground objects.

[0129] Optionally, the identification device 7 further includes:

[0130] a sixth processing unit, configured to perform fusion processing on the first sample image and the second sample image based on the PCA algorithm to obtain a plurality of sample fusion images; the first sample image and the second sample image are remote sensing images of the same geographical area at different times, and the first sample image and the second sample image contain different ground objects;

[0131] The seventh processing unit is used to pre-process the multiple sample fusion images, and train the multiple fusion-labeled images based on the target detection Cascade R-CNN model to obtain the preset new feature recognition model; wherein the fusion-labeled image is obtained by manually labeling the sample fusion image, and the manual labeling processing refers to the position labeling of the new features of the sample fusion image.

[0132] It should be noted that the above-mentioned embodiment provides a ground object recognition device that only uses the division of the above-mentioned functional modules as an example when executing the ground object recognition method. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the ground object recognition device provided in the above-mentioned embodiment and the ground object recognition method embodiment belong to the same concept, and the implementation process thereof is detailed in the method embodiment, which will not be repeated here.

[0133] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0134] The present application also provides a computer storage medium, which can store multiple instructions, which are suitable for being loaded by a processor and executing the above method steps. The specific execution process can be seen in Figure 2~Figure 6 The specific description of the illustrated embodiment will not be repeated here.

[0135] The present application also provides a terminal, comprising a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the above method steps.

[0136] See also Figure 8 , which shows a schematic diagram of the structure of a terminal involved in an embodiment of the present invention, and the terminal can be used to implement the method for identifying objects in the above-mentioned embodiment. Specifically:

[0137] The memory 803 can be used to store software programs and modules. The processor 800 executes various functional applications and data processing by running the software programs and modules stored in the memory 803. The memory 803 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the terminal device (such as audio data, a phone book, etc.), etc. In addition, the memory 803 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 803 may also include a memory controller to provide the processor 800 and the input unit 805 with access to the memory 803.

[0138] The input unit 805 can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control. Specifically, the input unit 805 may include a touch-sensitive surface 806 (e.g., a touch screen, a touch pad or a touch frame). The touch-sensitive surface 806, also known as a touch display screen or a touch pad, can collect user touch operations on or near it (e.g., operations performed by the user using any suitable object or accessory such as a finger, a stylus, etc. on or near the touch-sensitive surface 806), and drive the corresponding connection device according to a pre-set program. Optionally, the touch-sensitive surface 806 may include a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch direction, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into touch point coordinates, and then sends it to the processor 800, and can receive and execute commands sent by the processor 800. In addition, the touch-sensitive surface 806 can be implemented in various types such as resistive, capacitive, infrared and surface acoustic wave.

[0139] The display unit 813 can be used to display information input by the user or information provided to the user and various graphical user interfaces of the terminal device, which can be composed of graphics, text, icons, videos and any combination thereof. The display unit 813 may include a display panel 814. Optionally, the display panel 814 may be configured in the form of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), etc. Further, the touch-sensitive surface 806 may cover the display panel 814. When the touch-sensitive surface 806 detects a touch operation on or near it, it is transmitted to the processor 800 to determine the type of touch event. Subsequently, the processor 800 provides a corresponding visual output on the display panel 814 according to the type of touch event. Although in Figure 8 In the embodiment, the touch-sensitive surface 806 and the display panel 814 are implemented as two independent components to implement input and output functions, but in some embodiments, the touch-sensitive surface 806 and the display panel 814 can be integrated to implement input and output functions.

[0140] The processor 800 is the control center of the terminal device. It uses various interfaces and lines to connect various parts of the entire terminal device. By running or executing software programs and / or modules stored in the memory 803 and calling data stored in the memory 803, it executes various functions of the terminal device and processes data, thereby monitoring the terminal device as a whole. Optionally, the processor 800 may include one or more processing cores; wherein the processor 800 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application programs, etc., and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 800.

[0141] Specifically in this embodiment, the display unit of the terminal device is a touch screen display, and the terminal device also includes a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors. The one or more programs include steps to implement the above method.

[0142] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0143] In the various embodiments of the present invention, all functional units may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0144] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying a ground object, characterized in that: The method comprises: Acquire a third sample image, and perform cropping processing on the third sample image to obtain a plurality of third sample small-block images; wherein the third sample image refers to a remote sensing image containing a large number of ground objects and at a single moment; Preprocessing the plurality of third sample small-block images, and training the plurality of sample annotated images based on the target detection Cascade R-CNN model to obtain a preset ground object recognition model; wherein the sample annotated images are obtained by manually annotating the preprocessed third sample small-block images, and the sample annotated images contain location identifiers of the ground objects; The first sample image and the second sample image are fused based on the principal component analysis (PCA) algorithm to obtain a plurality of sample fused images; the first sample image and the second sample image are remote sensing images of the same geographical area at different times, and the first sample image and the second sample image contain different ground objects; Based on a preset object recognition model, multiple fused and annotated images are trained to obtain the preset newly added object recognition model; wherein the fused and annotated image is obtained by manually annotating the sample fused image, and the manual annotation processing refers to positionally annotating the newly added objects in the sample fused image; The first remote sensing image and the second remote sensing image are fused based on a principal component analysis (PCA) algorithm to obtain a fused image; wherein the first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times; The fused image is processed based on a preset new object recognition model to obtain a recognition result of the new object.

2. The method according to claim 1, characterized in that The first remote sensing image and the second remote sensing image contain different ground objects, and the recognition result of the newly added ground object is the position information of the newly added ground object in the second remote sensing image compared with the first remote sensing image.

3. The method according to claim 1, characterized in that The first remote sensing image and the second remote sensing image contain the same ground object, and the reminder information is displayed through the display unit.

4. The method according to claim 1, characterized in that Before fusing the first remote sensing image and the second remote sensing image to obtain a fused image based on the principal component analysis (PCA) algorithm, the method further includes: Acquire the first remote sensing image and the second remote sensing image; The fusing process of the first remote sensing image and the second remote sensing image based on the principal component analysis (PCA) algorithm to obtain a fused image includes: The first remote sensing image is cropped to obtain a plurality of first small-block images, and the second remote sensing image is cropped to obtain a plurality of second small-block images; The fused image is obtained by fusing the plurality of first small-block images and the plurality of second small-block images based on the PCA algorithm.

5. The method according to claim 4, characterized in that After the first remote sensing image is cropped to obtain a plurality of first small images, and the second remote sensing image is cropped to obtain a plurality of second small images, and before the plurality of first small images and the plurality of second small images are fused based on the PCA algorithm to obtain the fused image, the method further includes: The plurality of first small-block images and the plurality of second small-block images are preprocessed respectively; wherein the first small-block images correspond to the second small-block images one-to-one, and the preprocessing includes at least one of the following: stretching, brightness adjustment, and hue adjustment.

6. The method according to claim 1, characterized in that Before fusing the first sample image and the second sample image based on the PCA algorithm to obtain a plurality of sample fused images, the method further includes: acquiring the first sample image and the second sample image; The fusing process of the first sample image and the second sample image based on the PCA algorithm to obtain a plurality of sample fused images includes: Cutting the first sample image to obtain a plurality of first sample small-block images, and cutting the second sample image to obtain a plurality of second sample small-block images; The plurality of first sample small block images and the plurality of second sample small block images are preprocessed respectively, and the plurality of preprocessed first sample small block images and the plurality of preprocessed second sample small block images are fused based on the PCA algorithm to obtain the plurality of sample fused images.

7. The method according to claim 1, characterized in that Before the fused image is processed based on the preset new object recognition model to obtain the recognition result of the new object, the method further includes: Based on the PCA algorithm, a first sample image and a second sample image are fused to obtain a plurality of sample fused images; wherein the first sample image and the second sample image are remote sensing images of the same geographical area at different times, and the first sample image and the second sample image contain different ground objects; The multiple sample fused images are preprocessed, and multiple fused annotated images are trained based on the target detection Cascade R-CNN model to obtain the preset new feature recognition model; wherein the fused annotated images are obtained by manually annotating the sample fused images, and the manual annotation processing refers to the position annotation of the new features of the sample fused images.

8. A ground feature recognition device, characterized in that: The device comprises: A model training module is used to obtain a third sample image, and to crop the third sample image to obtain a plurality of third sample small-block images; wherein the third sample image refers to a remote sensing image containing a large number of objects and being a single moment; the plurality of third sample small-block images are preprocessed, and a plurality of sample annotated images are trained based on a target detection Cascade R-CNN model to obtain a preset object recognition model; wherein the sample annotated image is obtained by manually annotating the preprocessed third sample small-block image, and the sample annotated image contains the location identifier of the object; the first sample image and the second sample image are fused based on a principal component analysis (PCA) algorithm to obtain a plurality of sample fused images; the first sample image and the second sample image are remote sensing images of the same geographical area at different moments, and the first sample image and the second sample image contain different objects; the plurality of fused annotated images are trained based on a preset object recognition model to obtain the preset newly added object recognition model; wherein the fused annotated image is obtained by manually annotating the sample fused image, and the manual annotation process refers to the position annotation of the newly added objects in the sample fused image; A first processing module is used to fuse the first remote sensing image and the second remote sensing image based on a principal component analysis (PCA) algorithm to obtain a fused image; wherein the first remote sensing image and the second remote sensing image are remote sensing images of the same geographical area at different times; The second processing module is used to process the fused image based on a preset new feature recognition model to obtain a recognition result of the new feature.

9. A computer storage medium, characterized in that The computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the method steps according to any one of claims 1 to 7.

10. A terminal, characterized in that: include: A processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the method steps as claimed in any one of claims 1 to 7.

Images