Remote sensing image detection method and device based on deep neural network

Abstract

This invention provides a remote sensing image detection method and apparatus based on a deep neural network, relating to the field of remote sensing image detection technology. The method includes: acquiring a remote sensing image to be detected; segmenting the remote sensing image into multiple sub-images; each sub-image including at least one pixel; inputting each sub-image into a pre-trained image detection model to obtain the probability that each pixel in the at least one pixel of the sub-image belongs to a target land cover category among a set of preset land cover categories; the image detection model, based on a deep neural network, determines a first proportion of the region in each sub-image that matches the target land cover category based on the probability that each pixel in the at least one pixel belongs to the target land cover category among the preset land cover categories; and determining a second proportion of the region in the remote sensing image that matches the target land cover category based on the multiple first proportions of the multiple sub-images.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image detection technology, and more specifically, to a remote sensing image detection method and apparatus based on deep neural networks. Background Technology

[0002] In the field of remote sensing image detection technology, it is necessary to detect the content of various land cover categories in remote sensing images to further analyze geographical information such as vegetation type and topographic information. For most of the time, two-thirds of the Earth is covered by clouds, therefore, remote sensing images acquired by optical satellites are greatly affected by clouds. Furthermore, as a type of remote sensing image, the cloud and snow content of multispectral images serves as a criterion for judging the effective information content of the image. Remote sensing images with excessively high cloud and snow content will not provide sufficient information for subsequent use; therefore, it is necessary to detect the cloud and snow content in remote sensing images.

[0003] Currently, with multiple optical satellites operating in orbit simultaneously, these satellites upload a large amount of observation data and produce numerous high-resolution remote sensing images every day. However, manual inspection of these high-resolution remote sensing images is time-consuming and inefficient, failing to meet the requirements for rapid quality inspection of numerous remote sensing images.

[0004] With the explosive growth of machine learning technology in multiple fields, combining machine learning with remote sensing image detection tasks to develop automated, high-precision image detection models can effectively meet the needs of quality inspection for large batches of remote sensing images. In related technologies, to obtain more accurate land cover category detection results, high-resolution remote sensing images are directly inspected to achieve more precise detection results. However, this also significantly increases the computational load of the image detection model during the detection process, leading to a corresponding increase in computational resources consumed. Summary of the Invention

[0005] In view of this, the present invention provides a remote sensing image detection method and apparatus based on deep neural networks to solve the problem that when detecting land cover categories in high-resolution remote sensing images, the computational load of the image detection model increases significantly, and the computational resources consumed also increase accordingly.

[0006] A first aspect of the present invention provides a remote sensing image detection method based on a deep neural network, comprising: acquiring a remote sensing image to be detected; segmenting the remote sensing image into multiple sub-images; each sub-image including at least one pixel; inputting each sub-image into a pre-trained image detection model to obtain the probability that each pixel in the at least one pixel of the sub-image belongs to a target land cover category among a preset plurality of land cover categories; the image detection model being constructed based on a deep neural network; determining a first proportion of the region in each sub-image that conforms to the target land cover category in the sub-image based on the probability that each pixel in the at least one pixel belongs to the target land cover category among the preset plurality of land cover categories; and determining a second proportion of the region possessing the target land cover category in the remote sensing image based on the plurality of first proportions of the plurality of sub-images.

[0007] According to an embodiment of the present invention, the image detection model includes 2N convolutional modules, 2N first convolutional layers, N transposed convolutional modules, N second convolutional layers, and transposed convolutional layers; wherein, the j-th convolutional module is connected to the (j+1)-th convolutional module through the j-th first convolutional layer; the 2N-th convolutional module and the N-th transposed convolutional module are connected through the 2N-th first convolutional layer; the ith transposed convolutional module and the (i-1)-th transposed convolutional module are connected in series through the ith second convolutional layer; the first transposed convolutional module and the transposed convolutional layer are connected through the first second convolutional layer; wherein, j=1,2,…,2N-1; i=2,3,…,N.

[0008] According to an embodiment of the present invention, the j-th convolutional module and the 2N-th convolutional module are both used to perform convolution operations on the input data and output a first feature map; the j-th first convolutional layer is used to perform channel compression on the first feature map input by the j-th convolutional module, and use the compressed first feature map as the input of the (j+1)-th convolutional module; the 2N-th first convolutional layer is used to perform channel compression on the first feature map input by the 2N-th convolutional module, and use the compressed first feature map as the input of the N-th transposed convolutional module; the ith transposed convolutional module and the first transposed convolutional module are used to sequentially perform convolution operations on the input data. The system performs transposed convolution and convolution processing to output a second feature map. The i-th second convolutional layer is used to compress the channels of the second feature map input from the i-th transposed convolutional module, and uses the compressed second feature map as the input of the (i-1)-th transposed convolutional module. The first second convolutional layer is used to compress the channels of the second feature map input from the first transposed convolutional module, and uses the compressed second feature map as the input of the transposed convolutional layer. The transposed convolutional layer processes the input data and outputs the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among multiple preset land cover categories.

[0009] According to an embodiment of the present invention, in the image detection model, the (2i-1)th first convolutional layer is connected to the ith second convolutional layer; the ith first convolutional layer is connected to the ith second convolutional layer; the ith second convolutional layer is used to merge the input data of the (2i-1)th first convolutional layer and the ith transposed convolutional module, perform channel compression on the merged input data to obtain a compressed second feature map, and use the compressed second feature map as the input of the (i-1)th transposed convolutional module; the ith second convolutional layer is used to merge the input data of the ith first convolutional layer and the ith transposed convolutional module, perform channel compression on the merged input data to obtain a compressed second feature map, and use the compressed second feature map as the input of the transposed convolutional layer.

[0010] According to an embodiment of the present invention, the j-th first convolutional layer is further configured to sequentially perform convolution, normalization, and activation function operations on the data input to the j-th convolutional module to obtain a first feature map input to the (j+1)-th convolutional module; the i-th second convolutional layer is further configured to sequentially perform convolution, normalization, and activation function operations on the data input to the i-th transposed convolutional module to obtain a second feature map input to the (i-1)-th transposed convolutional module; the transposed convolutional layer is further configured to sequentially perform transposed convolution, normalization, and function operations on the input data to output the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among a plurality of preset land cover categories.

[0011] According to an embodiment of the present invention, any one of the 2N convolutional modules includes at least one convolutional kernel with a first step length; any one of the N transposed convolutional modules includes multiple convolutional kernels with a first step length and at least one transposed convolutional kernel with a second step length; wherein the first step length and the second step length are different.

[0012] According to an embodiment of the present invention, the i-th transposed convolution module is further configured to: perform a transposed convolution operation of a second step length on the input data to obtain a third feature map; perform a convolution operation of a first step length on the third feature map to obtain a fourth feature map; perform a convolution operation of a first step length on the fourth feature map to obtain a fifth feature map; and use the fifth feature map as the input of the (i-1)-th transposed convolution module.

[0013] According to an embodiment of the present invention, determining a second proportion of a region with a target land cover category in a remote sensing image based on a plurality of first proportions of a plurality of sub-images includes: acquiring a pre-constructed database, the database storing interrelated third proportions and proportion intervals; determining, in the database, a target proportion interval matching the first proportion of each sub-image, and a third proportion corresponding to the target proportion interval; and determining the second proportion of a region with a target land cover category in the remote sensing image based on a plurality of third proportions corresponding to a plurality of sub-images.

[0014] According to an embodiment of the present invention, the image detection model is trained by: acquiring historical remote sensing images containing multiple land cover categories; identifying land cover categories in the historical remote sensing images to generate a mask map of each land cover category in the historical remote sensing images; using the historical remote sensing images and the mask map of each land cover category as sample data, preprocessing the sample data to obtain training data; and training the pre-constructed image detection model with the training data.

[0015] A second aspect of the present invention provides a remote sensing image detection device based on a deep neural network, comprising: an image acquisition module for acquiring a remote sensing image to be detected and segmenting the remote sensing image into multiple sub-images; each sub-image includes at least one pixel;

[0016] The model processing module is used to input each sub-image into a pre-trained image detection model to obtain the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among multiple preset land cover categories; the image detection model is built based on a deep neural network.

[0017] The first proportion determination module is used to determine the first proportion of the region in each sub-image that matches the target land cover category based on the probability that each pixel in at least one pixel belongs to the target land cover category among a number of preset land cover categories; the second proportion determination module is used to determine the second proportion of the region that matches the target land cover category in the remote sensing image based on the multiple first proportions of the multiple sub-images.

[0018] A third aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.

[0019] A fourth aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the method described above.

[0020] The remote sensing image detection method and apparatus based on a deep neural network provided by the present invention can achieve the following technical effects:

[0021] The method in this embodiment divides the remote sensing image to be detected into multiple sub-images and detects the probability of each pixel in each sub-image belonging to a different land cover category. This reduces the computational load of the model when processing high-precision remote sensing images and improves the output rate. At the same time, based on the first proportion of multiple sub-images, the second proportion of the target land cover category in the remote sensing image is re-determined, further improving the accuracy of the region detection results for the target land cover category. Attached Figure Description

[0022] The above and other objects, features and advantages of the present invention will become more apparent from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0023] Figure 1 The flowchart illustrating a remote sensing image detection method based on a deep neural network according to an embodiment of the present invention is shown in the illustration.

[0024] Figure 2 One of the structural diagrams of an image detection model according to an embodiment of the present invention is illustrated schematically;

[0025] Figure 3 A second structural diagram of an image detection model according to an embodiment of the present invention is shown schematically.

[0026] Figure 4A A schematic diagram illustrating the structure of a first convolutional layer according to an embodiment of the present invention is shown.

[0027] Figure 4B A schematic diagram illustrating the structure of a second convolutional layer according to an embodiment of the present invention is shown.

[0028] Figure 4C A schematic diagram of the structure of a transposed convolutional layer according to an embodiment of the present invention is shown.

[0029] Figures 5A to 5D The structural diagrams of four convolutional modules according to embodiments of the present invention are illustrated respectively;

[0030] Figure 6 A schematic diagram of the transposed convolution module according to an embodiment of the present invention is shown.

[0031] Figure 7 A flowchart illustrating a training method for an image detection module according to an embodiment of the present invention is shown schematically.

[0032] Figure 8 A block diagram of a remote sensing image detection device based on a deep neural network according to an embodiment of the present invention is shown schematically.

[0033] Figure 9 A block diagram illustrating an embodiment of the present invention suitable for implementing the electronic device described above is shown. Detailed Implementation

[0034] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0035] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0036] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0037] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0038] In this embodiment of the invention, the remote sensing image to be detected can be a multispectral image with various different land cover categories, such as mountains, fields, cities, wetlands, water bodies, oceans, snowfields, clouds, and forests. In this embodiment, excessively high cloud and snow content can prevent the remote sensing image from providing sufficient information for subsequent use. Therefore, identifying the proportion of snowfield and cloud cover areas in the remote sensing image can support users in understanding the effective information content of the remote sensing image.

[0039] It should be noted that the embodiments of the present invention only provide illustrative application scenarios for detecting land cover categories in remote sensing images, and are not intended to limit the present invention.

[0040] Figure 1 The flowchart illustrating a remote sensing image detection method based on a deep neural network according to an embodiment of the present invention is shown.

[0041] like Figure 1As shown, the remote sensing image detection method based on deep neural networks in this embodiment includes operations S101 to S104.

[0042] In operation S101, the remote sensing image to be detected is acquired and the remote sensing image is segmented into multiple sub-images; each sub-image includes at least one pixel.

[0043] For example, the remote sensing image to be detected can be an 8000×8000 resolution remote sensing image, which can be divided into multiple sub-images with a resolution of 400×400.

[0044] In operation S102, each sub-image is input into a pre-trained image detection model to obtain the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among multiple preset land cover categories; the image detection model is constructed based on a deep neural network.

[0045] In this embodiment of the invention, each sub-image is input into a pre-trained image detection model. The probability that each pixel in each sub-image belongs to a target land cover category among a set of preset land cover categories is obtained. For example, the preset land cover categories include ocean, snow, clouds, and forest. Accordingly, for any pixel in a sub-image, the probability that the pixel belongs to the ocean, snow, cloud, or forest category can be obtained respectively. The image detection model in this embodiment of the invention is based on a deep neural network.

[0046] In operation S103, based on the probability that each pixel in at least one pixel belongs to the target land cover category among a plurality of preset land cover categories, the first proportion of the region in each sub-image that conforms to the target land cover category is determined.

[0047] For example, snow and clouds are used as target land cover categories. In each sub-image, the first proportion of areas matching the target land cover category is determined. For example, for a pixel in a sub-image, the probability of that pixel belonging to the snow category and the cloud category is 10% and 15%, respectively. The probability of each pixel belonging to the snow and cloud categories is iterated through in the sub-image. The sum of the probabilities of each pixel belonging to the snow category is obtained, and this sum is divided by the number of pixels in the sub-image to obtain the proportion of snow areas in the sub-image. Similarly, the proportion of cloud areas in the sub-image is obtained. The sum of the proportions of snow areas and cloud areas in the sub-image is used as the first proportion.

[0048] In operation S104, based on multiple first proportions of multiple sub-images, a second proportion of the region with the target land cover category in the remote sensing image is determined.

[0049] The method in this embodiment divides the remote sensing image to be detected into multiple sub-images and detects the probability of each pixel in each sub-image belonging to a different land cover category. This reduces the computational load of the model when processing high-precision remote sensing images and improves the output rate. At the same time, based on the first proportion of multiple sub-images, the second proportion of the target land cover category in the remote sensing image is re-determined, further improving the accuracy of the region detection results for the target land cover category.

[0050] Furthermore, the above operation S104, which determines the second proportion of the region with the target land cover category in the remote sensing image based on the multiple first proportions of multiple sub-images, may also include: acquiring a pre-constructed database, which stores interrelated third proportions and proportion intervals; determining, in the database, the target proportion interval matching the first proportion of each sub-image, and the third proportion corresponding to the target proportion interval; and determining the second proportion of the region with the target land cover category in the remote sensing image based on the multiple third proportions corresponding to the multiple sub-images.

[0051] To make the cloud cover calculation results of the image detection model more closely resemble the subjective identification results of quality inspectors, a pre-built database can be constructed. This database obtains the third proportion of target land cover categories in historical remote sensing images obtained by quality inspectors through subjective judgment. It should be noted that due to the differences between human visual recognition and machine learning processes, for the same remote sensing image, when the human eye identifies the proportion of clouds, a sparse cloud area will be considered entirely clouded; while machine recognition, accurate to the pixel level, will consider a sparse cloud area as having a certain degree (e.g., 60%, depending on the degree of cloud sparseness). Therefore, to ensure that multiple first proportions of multiple sub-images conform to human judgment, it is necessary to construct interrelated third proportions and proportion intervals. Specifically, all intervals from 0 to 1 are divided into multiple proportion intervals; each proportion interval corresponds to a third proportion; and each proportion interval also corresponds to a first proportion.

[0052] Furthermore, a target proportion range matching the first proportion of each sub-image is determined, along with a third proportion corresponding to the target proportion range. Based on multiple third proportions corresponding to multiple sub-images, the second proportion of the region possessing the target land cover category in the remote sensing image is determined by averaging the multiple third proportions.

[0053] Figure 2 One of the structural diagrams of an image detection model according to an embodiment of the present invention is illustrated schematically.

[0054] like Figure 2As shown, the image detection model includes 2N convolutional modules, 2N first convolutional layers, N transposed convolutional modules, N second convolutional layers, and transposed convolutional layers. Specifically, the j-th convolutional module is connected to the (j+1)-th convolutional module through the j-th first convolutional layer; the 2N-th convolutional module and the N-th transposed convolutional module are connected through the 2N-th first convolutional layer; the ith transposed convolutional module and the (i-1)-th transposed convolutional module are connected in series through the ith second convolutional layer; and the first transposed convolutional module and the transposed convolutional layer are connected through the first second convolutional layer. Here, j = 1, 2, ..., 2N-1; i = 2, 3, ..., N.

[0055] For example, the image detection model includes eight convolutional modules, eight first convolutional layers, four transposed convolutional modules, four second convolutional layers, and one transposed convolutional layer. Figure 2 As shown on the left, the convolutional modules from top to bottom are the first to the eighth convolutional modules; the convolutional layers from top to bottom are the first to the eighth convolutional layers. Figure 2 As shown on the right, the transposed convolutional modules from top to bottom are the first transposed convolutional module to the fourth transposed convolutional module; the convolutional layers from top to bottom are the first second convolutional layer to the fourth second convolutional layer; wherein, the eighth convolutional module and the fourth transposed convolutional module are connected through the eighth first convolutional layer.

[0056] For example, an image detection model can have eight levels of convolutional modules and four levels of transposed convolutional modules. It should be noted that the number of levels of the convolutional and transposed convolutional modules in this application can be specifically configured according to the image resolution of the remote sensing image to be detected. The number of levels of the convolutional and transposed convolutional modules can increase as the resolution of the remote sensing image increases. Higher-level image detection models can produce more accurate output results, but they also increase the computational load of the image detection module.

[0057] exist Figure 2 The image detection model shown is designed for detecting cloud cover and snow cover in remote sensing images. By configuring eight convolutional modules and four transposed convolutional modules, as well as first and second convolutional layers corresponding to the number of layers, the model can ensure the accuracy of the output while avoiding excessive computation due to too many layers.

[0058] The following describes the embodiments of the present invention. Figure 2 The functions of each convolutional module, transposed convolutional module, first convolutional layer, second convolutional layer, and transposed convolutional layer shown are explained.

[0059] In this embodiment of the invention, the j-th convolutional module and the 2N-th convolutional module are both used to perform convolution operations on the input data and output the first feature map.

[0060] The j-th first convolutional layer is used to perform channel compression on the first feature map input to the j-th convolutional module, and use the compressed first feature map as the input to the (j+1)-th convolutional module; the 2N-th first convolutional layer is used to perform channel compression on the first feature map input to the 2N-th convolutional module, and use the compressed first feature map as the input to the N-th transposed convolutional module.

[0061] The i-th transposed convolutional module and the first transposed convolutional module are used to perform transposed convolution and convolution processing on the input data in sequence, and output the second feature map.

[0062] The i-th second convolutional layer is used to compress the second feature map input from the i-th transposed convolutional module through channels, and the compressed second feature map is used as the input of the (i-1)-th transposed convolutional module; the 1-th second convolutional layer is used to compress the second feature map input from the 1-th transposed convolutional module through channels, and the compressed second feature map is used as the input of the transposed convolutional layer.

[0063] The transposed convolutional layer is used to process the input data and output the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among a number of preset land cover categories.

[0064] Figure 3 The second schematic diagram illustrates the structure of an image detection model according to an embodiment of the present invention.

[0065] like Figure 3 As shown, in the image detection model, the 2i-1th first convolutional layer is connected to the 1st second convolutional layer; the 1st first convolutional layer is connected to the 1st second convolutional layer.

[0066] The convolutional layers from top to bottom are as follows: the first convolutional layer is connected to the first convolutional layer; the third convolutional layer is connected to the second convolutional layer; the fifth convolutional layer is connected to the third convolutional layer; and the seventh convolutional layer is connected to the fourth convolutional layer.

[0067] The following describes the embodiments of the present invention. Figure 3 The functions of the four cascaded second convolutional layers shown are explained.

[0068] In the second convolutional layer provided in this embodiment of the invention, the i-th second convolutional layer is used to merge the input data of the (2i-1)-th first convolutional layer and the i-th transposed convolutional module, perform channel compression on the merged input data to obtain a compressed second feature map, and use the compressed second feature map as the input of the (i-1)-th transposed convolutional module. The first second convolutional layer is used to merge the input data of the first first convolutional layer and the first transposed convolutional module, perform channel compression on the merged input data to obtain a compressed second feature map, and use the compressed second feature map as the input of the transposed convolutional layer.

[0069] In embodiments of the present invention, such as Figure 3 As shown, the convolutional module and the first convolutional layer on the left constitute the encoder. The transposed convolutional module and the second convolutional layer on the right construct the decoder. The encoder is used to extract spatial feature data from the sub-image, reducing the spatial dimension while increasing the number of channels. The decoder performs projection mapping on the extracted data to create a mask map containing the cloud and snow detection results at the original resolution of the sub-image.

[0070] Furthermore, such as Figure 3 As shown, a cascaded approach is used to connect feature maps from the encoding and decoding processes to achieve fusion between low-level and high-level features. The purpose of the first and second convolutional layers is to extract translation-invariant features from the image. In each convolutional layer, the convolutional kernel is convolved with the input image to generate the corresponding feature map. Compared to using fully connected layers, in this embodiment, the convolutional kernels in the first and second convolutional layers share the same weights across the entire sub-image, effectively reducing the weight parameters of the image detection model and learning translation-invariant features on the sub-image.

[0071] Figure 4A This is a structural diagram of the first convolutional layer provided in an embodiment of the present invention.

[0072] like Figure 4A As shown, the j-th first convolutional layer is also used to sequentially perform convolution operations, normalization operations, and activation function operations on the data input to the j-th convolutional module to obtain the first feature map input to the (j+1)-th convolutional module.

[0073] Figure 4B This is a structural diagram of the second convolutional layer provided in an embodiment of the present invention.

[0074] like Figure 4B As shown, the i-th second convolutional layer is also used to perform convolution or transpose convolution operations, normalization operations and activation function operations on the data input to the i-th transpose convolutional module in sequence, so as to obtain the second feature map input to the (i-1)-th transpose convolutional module.

[0075] In this embodiment of the invention, the first convolutional layer and the second convolutional layer can be configured with the same structure, or they can be configured with other different structures. Figure 4A and Figure 4B The structures of the first and second convolutional layers in this invention are given only as examples and are not intended to limit the structure of the two convolutional layers.

[0076] Figure 4C This is a structural diagram of the transposed convolutional layer provided in an embodiment of the present invention.

[0077] like Figure 4C As shown, the transposed convolutional layer is also used to sequentially perform transposed convolution, normalization, and function operations on the input data, and output the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among multiple preset land cover categories.

[0078] For example, the activation function in the embodiments of the present invention can be configured as a ReLU function.

[0079] The structure of the 2N convolutional modules and the N transposed convolutional modules in the embodiments of the present invention will be described below.

[0080] In this embodiment of the invention, any one of the 2N convolutional modules includes at least one convolutional kernel with a first step length; any one of the N transposed convolutional modules includes multiple convolutional kernels with a first step length and at least one transposed convolutional kernel with a second step length; wherein the first step length and the second step length are different.

[0081] The structure of each convolution module in the embodiments of the present invention will be described below.

[0082] For example, the eight convolutional modules have four structural types: A, B, C, and D. The first, third, fifth, and seventh convolutional modules are of type A, C, and D, respectively, while the second, fourth, sixth, and eighth convolutional modules are all of type B. The four transposed convolutional modules have the same structure.

[0083] like Figure 5A The diagram shown is a structural diagram of a convolution module of type A provided in an embodiment of the present invention.

[0084] like Figure 5B The diagram shown is a structural diagram of a type B convolutional module provided in an embodiment of the present invention.

[0085] like Figure 5C The diagram shown is a structural diagram of a type C convolution module provided in an embodiment of the present invention.

[0086] like Figure 5DThe diagram shown is a structural diagram of a type D convolution module provided in an embodiment of the present invention.

[0087] exist Figures 5A-5D In the first, third, fifth, and seventh convolutional modules, the stride of the convolutional kernels is 1.

[0088] exist Figures 5A to 5D In the model, the second, fourth, sixth, and eighth convolutional modules, in addition to a convolutional kernel with a stride of 1, also include two convolutional kernels with a stride of 2. By using convolutions with a stride of 2 to downsample the sub-image, the size of the input image data is reduced and the dimensionality of the input is increased, resulting in a more compact feature map. The second, fourth, sixth, and eighth convolutional modules correspond to four downsampling operations, each with a different receptive field: pixel-level, small window around each pixel, larger window around each pixel, and the maximum visual field of the model.

[0089] Figure 6 A schematic diagram of the transposed convolution module according to an embodiment of the present invention is shown.

[0090] like Figure 6 As shown, in the transposed convolution module provided in this embodiment of the invention, the i-th transposed convolution module is used to perform a transposed convolution operation with a second step length on the input data to obtain a third feature map; perform a convolution operation with a first step length on the third feature map to obtain a fourth feature map; perform a convolution operation with a first step length on the fourth feature map to obtain a fifth feature map; and use the fifth feature map as the input of the (i-1)-th transposed convolution module.

[0091] The first transposed convolution module performs a second-stride transposed convolution operation on the input data to obtain the third feature map; a first-stride convolution operation is performed on the third feature map to obtain the fourth feature map; a first-stride convolution operation is performed on the fourth feature map to obtain the fifth feature map; and the fifth feature map is used as the input to the transposed convolution module.

[0092] The process of image detection module processing sub-images in this embodiment of the invention will be described below.

[0093] First, a 400×400×m sub-image is input into the image detection model and passed through convolution module A; where m is the number of channels.

[0094] In convolution module A, the input image is convolved with a stride of 1 using a 3×3 convolution to obtain feature map a1; convolution processing is performed within the first convolutional layer, the result is normalized, and then the ReLU activation function is applied; the image is convolved twice with a stride of 1 using a 3×3 convolution to obtain feature map a2; the image is convolved three times with a stride of 1 using a 3×3 convolution to obtain feature map a3; the input image is merged with the channels of a1, a2, and a3 to obtain A1; A1 is convolved with a stride of 1 using a 1×1 convolution and channel compression is performed to obtain a 400×400×m feature map. Figure 1 ; Features Figure 1 The input is fed into convolution module B.

[0095] In convolution module B, a 3×3 convolution with a stride of 2 is performed to obtain feature map b1_1;

[0096] Features Figure 1 Perform a 3×3 convolution with a stride of 1, followed by a 3×3 convolution with a stride of 2 to obtain feature map b1_2; merge the channels of feature map b1_1 and feature map b1_2 to obtain B1; perform a 1×1 convolution with a stride of 1 on B1 and compress the channels to obtain a 200×200×m feature map. Figure 2 ; Features Figure 2 The input is fed into convolution module C.

[0097] In convolution module C, a 3×3 convolution with a stride of 1 is performed to obtain feature map c2_1; for the feature... Figure 2 Perform two 3×3 convolutions with a stride of 1 to obtain feature map c2_2; then... Figure 2 C2 is obtained by merging channels of c2_1 and c2_2; C2 is then subjected to a 1×1 convolution with a stride of 1, followed by channel compression to obtain a 200×200×m feature. Figure 3 ; Features Figure 3 The input is fed into convolution module B.

[0098] In convolution module B, a 3×3 convolution with a stride of 2 is performed to obtain feature map b3_1; for the feature... Figure 3 Perform a 3×3 convolution with a stride of 1, then perform a 3×3 convolution with a stride of 2 to obtain feature map b3_2; merge the channels of feature map b3_1 and feature map b3_2 to obtain B3; perform a 1×1 convolution with a stride of 1 on B3 and compress the channels to obtain a 100×100×m feature map 4; input feature map 4 into convolution module C.

[0099] In convolution module C, a 3×3 convolution with a stride of 1 is performed to obtain feature map c4_1; feature map c4_1 is then subjected to two 3×3 convolutions with a stride of 1 to obtain feature map c4_2; feature map c4_1 is then merged with c4_1 and c4_2 to obtain C4; C4 is then subjected to a 1×1 convolution with a stride of 1 and channel compression is performed to obtain feature map 5, which is 100×100×m; feature map 5 is then input into convolution module B.

[0100] In convolution module B, a 3×3 convolution with a stride of 2 is performed to obtain feature map b5_1; a 3×3 convolution with a stride of 1 is performed on feature map b5_1, followed by a 3×3 convolution with a stride of 2 to obtain feature map b5_2; feature maps b5_1 and b5_2 are merged by channel merging to obtain B5; B5 is then subjected to a 1×1 convolution with a stride of 1 and channel compression to obtain a 50×50×m feature map. Figure 6 ; Features Figure 6 The input is fed into convolution module D.

[0101] In convolution module D, a 1×3 convolution with a stride of 1 is performed, followed by a 3×1 convolution with a stride of 1, to obtain feature map d6_1; for the features... Figure 6 Perform a 1×5 convolution with a stride of 1, followed by a 5×1 convolution with a stride of 1 to obtain feature map d6_2; [The text abruptly ends here, likely due to an incomplete sentence or missing information.] Figure 6 Perform a 1×7 convolution with a stride of 1, and then perform a 7×1 convolution with a stride of 1 to obtain the feature map d6_3;

[0102] Features Figure 6 Channel merging is performed with d6_1, d6_2, and d6_3 to obtain D6; a 1×1 convolution with a stride of 1 is performed on D6, followed by channel compression to obtain a 50×50×m feature. Figure 7 ; Features Figure 7 The input is fed into convolution module B.

[0103] In convolution module B, a 3×3 convolution with a stride of 2 is performed to obtain feature map b7_1; a 3×3 convolution with a stride of 1 is performed on feature map b7_1, followed by a 3×3 convolution with a stride of 2 to obtain feature map b7_2; feature maps b7_1 and b7_2 are merged by channel merging to obtain B7; B7 is then subjected to a 1×1 convolution with a stride of 1 and channel compression to obtain a 25×25×m feature map. Figure 8 ; Features Figure 8 The input is fed into the fourth transposed convolution module.

[0104] In the fourth transposed convolution module, a 3×3 transposed convolution with a stride of 2 is performed to obtain feature map u8_1; a 3×3 convolution with a stride of 1 is performed on feature map u8_1 to obtain feature map u8_2; a 3×3 convolution with a stride of 1 is performed on feature map u8_2 to obtain feature map u8_3; feature map u8_3 is then convolved with the feature map u8_1. Figure 7 Concatenate the features to obtain feature map U8; perform a 1×1 convolution with a stride of 1 on feature map U8 to obtain a 50×50×m feature map. Figure 9 ; Features Figure 9 The input is fed into the transpose convolution module.

[0105] In the third transposed convolution module, a 3×3 transposed convolution with a stride of 2 is performed to obtain feature map u9_1; transposed convolution is performed within the third transposed convolution layer, the result is normalized, and then the ReLU activation function is applied; a 3×3 convolution with a stride of 1 is performed on feature map u9_1 to obtain feature map u9_2; a 3×3 convolution with a stride of 1 is performed on feature map u9_2 to obtain feature map u9_3; feature map u9_3 is concatenated with feature map 5 to obtain feature map U9; a 1×1 convolution with a stride of 1 is performed on feature map U9 to obtain 100×100×m feature map 10; feature map 10 is then input into the transposed convolution module.

[0106] In the second transposed convolution module, a 3×3 transposed convolution with a stride of 2 is performed to obtain feature map u10_1; a 3×3 convolution with a stride of 1 is performed on feature map u10_1 to obtain feature map u10_2; a 3×3 convolution with a stride of 1 is performed on feature map u10_2 to obtain feature map u10_3; feature map u10_3 is then convolved with the feature map u10_3. Figure 3 Concatenate the features to obtain feature map U10; perform a 1×1 convolution with a stride of 1 on feature map U10 to obtain feature map 11 of 200×200×m; input feature map 11 into the transposed convolution module.

[0107] In the first transposed convolution module, a 3×3 transposed convolution with a stride of 2 is performed to obtain feature map u11_1; a 3×3 convolution with a stride of 1 is performed on feature map u11_1 to obtain feature map u11_2; a 3×3 convolution with a stride of 1 is performed on feature map u11_2 to obtain feature map u11_3; feature map u11_3 is then convolved with the feature map u11_2. Figure 1 Concatenate the layers to obtain feature map U11; perform a 1×1 convolution with stride 1 on feature map U11 to obtain feature map 12 of 400×400×m; input feature map 12 into the transposed convolutional layer.

[0108] Next, in the transposed convolutional layer, a 1×1 transposed convolution with a stride of 1 is performed to obtain the feature values.

[0109] Finally, using Softmax as the activation function, the feature values ​​are input into the Softmax function to calculate the first proportion of the region in each sub-image that matches the target land cover category in the sub-image, and then operations S103 to S104 are performed.

[0110] Figure 7 A flowchart illustrating a training method for an image detection module according to an embodiment of the present invention is shown.

[0111] like Figure 7 As shown, the training method of the image detection module in this embodiment includes operations S701 to S703.

[0112] In this embodiment of the invention, the image detection model is trained in the following manner:

[0113] Operation S701: Acquire historical remote sensing images containing the multiple land cover categories.

[0114] In step S702, the land cover category is identified in the historical remote sensing image, and a mask image for each land cover category in the historical remote sensing image is generated.

[0115] In step S703, the historical remote sensing images and the mask images of each land cover category are used as sample data. The sample data is preprocessed to obtain training data. The pre-built image detection model is then trained using the training data.

[0116] Based on the remote sensing image detection method based on deep neural networks, this invention also provides a remote sensing image detection device based on deep neural networks. The following will combine... Figure 8 The device is described in detail.

[0117] Figure 8 A block diagram of a remote sensing image detection apparatus based on a deep neural network according to an embodiment of the present invention is shown schematically.

[0118] like Figure 8 As shown, the remote sensing image detection device 800 based on deep neural networks in this embodiment includes an image acquisition module 810, a model processing module 820, a first proportion determination module 830, and a second proportion determination module 840.

[0119] The image acquisition module 810 is used to acquire the remote sensing image to be detected and to segment the remote sensing image into multiple sub-images; each sub-image includes at least one pixel.

[0120] The model processing module 820 is used to input each sub-image into a pre-trained image detection model to obtain the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among a number of preset land cover categories; the image detection model is constructed based on a deep neural network.

[0121] The first proportion determination module 830 is used to determine the first proportion of the area in each sub-image that conforms to the target land cover category based on the probability that each pixel in at least one pixel belongs to the target land cover category among a plurality of preset land cover categories.

[0122] The second proportion determination module 840 is used to determine the second proportion of the region with the target land cover category in the remote sensing image based on the multiple first proportions of multiple sub-images.

[0123] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present invention, or at least part of the functions of any one or more of them, can be implemented in a single module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be implemented by being divided into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, and firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0124] For example, any plurality of the image acquisition module 810, model processing module 820, first proportion determination module 830, and second proportion determination module 840 can be combined into one module / unit / subunit, or any one of these modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least part of the functionality of one or more of these modules / units / subunits can be combined with at least part of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of the present invention, at least one of the image acquisition module 810, model processing module 820, first proportion determination module 830, and second proportion determination module 840 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the image acquisition module 810, model processing module 820, first proportion determination module 830 and second proportion determination module 840 may be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0125] It should be noted that the remote sensing image detection device based on deep neural networks in the embodiments of the present invention corresponds to the remote sensing image detection method based on deep neural networks in the embodiments of the present invention, and their specific implementation details are the same, so they will not be repeated here.

[0126] Figure 9 A block diagram of an electronic device suitable for implementing the methods described above, according to an embodiment of the present invention, is shown schematically. Figure 9 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0127] like Figure 9As shown, an electronic device 900 according to an embodiment of the present invention includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory configured for caching purposes. The processor 901 may include a single processing unit or multiple processing units configured to perform different actions of the method flow according to an embodiment of the present invention.

[0128] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 902 and / or RAM 903. It should be noted that the programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0129] According to an embodiment of the present invention, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0130] According to embodiments of the present invention, the method flow according to embodiments of the present invention can be implemented as a computer software program. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code configured to perform the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 909, and / or installed from removable medium 911. When the computer program is executed by processor 901, it performs the functions defined in the system of the embodiments of the present invention. According to embodiments of the present invention, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0131] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0132] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0133] For example, according to embodiments of the present invention, a computer-readable storage medium may include one or more memories other than the ROM 902 and / or RAM 903 described above.

[0134] Embodiments of the present invention also include a computer program product comprising a computer program containing program code configured to perform the methods provided in the embodiments of the present invention. When the computer program product is run on an electronic device, the program code is configured to enable the electronic device to implement the remote sensing image detection method based on a deep neural network provided in the embodiments of the present invention.

[0135] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this embodiment of the invention. According to embodiments of the invention, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0136] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0137] According to embodiments of the present invention, program code configured to execute the computer programs provided in the embodiments of the present invention can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0138] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions configured to perform a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or pairings fall within the scope of this invention.

[0139] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A remote sensing image detection method based on deep neural networks, comprising: Acquire the remote sensing image to be detected, and segment the remote sensing image into multiple sub-images; Each of the sub-images includes at least one pixel; Each of the sub-images is input into a pre-trained image detection model to obtain the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among a set of multiple land cover categories; the image detection model is constructed based on a deep neural network. Based on the probability that each pixel in the at least one pixel belongs to the target land cover category among a preset plurality of land cover categories, determine the first proportion of the area in each sub-image that conforms to the target land cover category in the sub-image; Based on multiple first proportions of the multiple sub-images, a second proportion of the region possessing the target land cover category in the remote sensing image is determined; The image detection model includes 2N convolutional modules, 2N first convolutional layers, N transposed convolutional modules, N second convolutional layers, and transposed convolutional layers; In this configuration, the j-th convolutional module is connected to the (j+1)-th convolutional module through the j-th first convolutional layer; the 2N-th convolutional module and the N-th transposed convolutional module are connected through the 2N-th first convolutional layer; the ith transposed convolutional module and the (i-1)-th transposed convolutional module are connected in series through the ith second convolutional layer; and the 1st transposed convolutional module is connected to the transposed convolutional layer through the 1st second convolutional layer; where j = 1, 2, ..., 2N-1; i = 2, 3, ..., N; Wherein, the (2i-1)th first convolutional layer is connected to the ith second convolutional layer; the ith first convolutional layer is connected to the ith second convolutional layer; the ith second convolutional layer is used to merge the input data of the (2i-1)th first convolutional layer and the ith transposed convolutional module, perform channel compression on the merged input data to obtain a compressed second feature map, and use the compressed second feature map as the input of the (i-1)th transposed convolutional module; the ith second convolutional layer is used to merge the input data of the ith first convolutional layer and the ith transposed convolutional module, perform channel compression on the merged input data to obtain a compressed second feature map, and use the compressed second feature map as the input of the transposed convolutional layer.

2. The method according to claim 1, wherein the j-th convolutional module and the 2N-th convolutional module are both used to perform convolution operations on the input data and output a first feature map; The j-th first convolutional layer is used to perform channel compression on the first feature map input to the j-th convolutional module, and use the compressed first feature map as the input to the (j+1)-th convolutional module; the 2N-th first convolutional layer is used to perform channel compression on the first feature map input to the 2N-th convolutional module, and use the compressed first feature map as the input to the N-th transposed convolutional module. The i-th transposed convolution module and the 1st transposed convolution module are used to sequentially perform transposed convolution processing and convolution processing on the input data to output a second feature map; The i-th second convolutional layer is used to perform channel compression on the second feature map input from the i-th transposed convolutional module, and use the compressed second feature map as the input of the (i-1)-th transposed convolutional module; the first second convolutional layer is used to perform channel compression on the second feature map input from the first transposed convolutional module, and use the compressed second feature map as the input of the transposed convolutional layer. The transposed convolutional layer is used to process the input data and output the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among a plurality of preset land cover categories.

3. According to the method of claim 2, the j-th first convolutional layer is further configured to sequentially perform convolution operation, normalization operation and activation function operation on the data input to the j-th convolutional module to obtain the first feature map input to the (j+1)-th convolutional module; The i-th second convolutional layer is also used to perform convolution, normalization and activation function operations on the data input to the i-th transposed convolutional module in sequence to obtain the second feature map input to the (i-1)-th transposed convolutional module; The transposed convolutional layer is also used to sequentially perform transposed convolution, normalization, and function operations on the input data, and output the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among a plurality of preset land cover categories.

4. The method according to claim 2, wherein any one of the 2N convolutional modules comprises at least one convolutional kernel with a first-step length; Any of the N transposed convolutional modules includes multiple convolutional kernels with a first step length and at least one transposed convolutional kernel with a second step length. in, The first step length and the second step length are different.

5. The method according to claim 4, wherein the i-th transposed convolutional module is further configured to: Perform a transpose convolution operation of the second step size on the input data to obtain the third feature map; Perform the first-step convolution operation on the third feature map to obtain the fourth feature map; Perform the first-step convolution operation on the fourth feature map to obtain the fifth feature map; The fifth feature map is used as the input to the (i-1)th transposed convolutional module.

6. The method according to claim 1, wherein determining a second proportion of the region possessing the target land cover category in the remote sensing image based on a plurality of first proportions of the plurality of sub-images comprises: Obtain a pre-built database that stores interrelated third proportions and proportion ranges; In the database, a target proportion range that matches the first proportion of each sub-image is determined, and a third proportion corresponding to the target proportion range is determined; Based on the multiple third proportions corresponding to the multiple sub-images, a second proportion of the region possessing the target land cover category in the remote sensing image is determined.

7. The method according to claim 1, wherein the image detection model is trained in the following manner: Acquire historical remote sensing images containing the multiple land cover categories; The historical remote sensing images are used to identify land cover categories, and a mask image for each land cover category in the historical remote sensing images is generated. The historical remote sensing images and the mask images of each land cover category are used as sample data. The sample data is preprocessed to obtain training data. The pre-built image detection model is trained using the training data.

8. A remote sensing image detection device based on a deep neural network, comprising: The image acquisition module is used to acquire the remote sensing image to be detected and to segment the remote sensing image into multiple sub-images; Each of the sub-images includes at least one pixel; The model processing module is used to input each of the sub-images into a pre-trained image detection model to obtain the probability that each pixel in at least one pixel of the sub-image belongs to the target land cover category among a set of multiple land cover categories; the image detection model is constructed based on a deep neural network. The first proportion determination module is used to determine the first proportion of the region in each sub-image that conforms to the target land cover category in the sub-image based on the probability that each pixel in the at least one pixel belongs to the target land cover category among a preset plurality of land cover categories; The second proportion determination module is used to determine the second proportion of the region with the target land cover category in the remote sensing image based on the multiple first proportions of the multiple sub-images; The image detection model includes 2N convolutional modules, 2N first convolutional layers, N transposed convolutional modules, N second convolutional layers, and transposed convolutional layers; In this configuration, the j-th convolutional module is connected to the (j+1)-th convolutional module through the j-th first convolutional layer; the 2N-th convolutional module and the N-th transposed convolutional module are connected through the 2N-th first convolutional layer; the ith transposed convolutional module and the (i-1)-th transposed convolutional module are connected in series through the ith second convolutional layer; and the 1st transposed convolutional module is connected to the transposed convolutional layer through the 1st second convolutional layer; where j = 1, 2, ..., 2N-1; i = 2, 3, ..., N; Wherein, the (2i-1)th first convolutional layer is connected to the ith second convolutional layer; the ith first convolutional layer is connected to the ith second convolutional layer; the ith second convolutional layer is used to merge the input data of the (2i-1)th first convolutional layer and the ith transposed convolutional module, perform channel compression on the merged input data to obtain a compressed second feature map, and use the compressed second feature map as the input of the (i-1)th transposed convolutional module; the ith second convolutional layer is used to merge the input data of the ith first convolutional layer and the ith transposed convolutional module, perform channel compression on the merged input data to obtain a compressed second feature map, and use the compressed second feature map as the input of the transposed convolutional layer.

Images