Sar image change detection method and apparatus with reduced interference

Abstract

The application relates to the technical field of photogrammetry, in particular to a SAR image change detection method and device for reducing interference, wherein the method comprises the following steps: identifying each pixel point of a to-be-detected SAR image, and generating a patch of a target size based on the center of each pixel point; inputting the patch into a preset encoding model to obtain an encoded feature map with an interference degree reduced to a target interference degree; inputting the encoded feature map into a preset decoding model to generate a cascaded feature map, and generating a change detection result of the SAR image based on the cascaded feature map. According to the application, the encoding model can be used to convert the feature map of the to-be-detected SAR image from a spatial domain to a frequency domain, thereby separating and encoding the high-frequency noise of the feature map, obtaining spatial information and position information, reducing the influence of the noise on the change detection, and then the decoding model can be used to bridge the multi-scale features and convert the feature map from the frequency domain back to the spatial domain, so that the precision of the SAR image change detection is effectively improved.

Description

Technical Field

[0001] This application relates to the field of photogrammetry technology, and in particular to a method and apparatus for detecting changes in SAR images with reduced interference. Background Technology

[0002] In related technologies, SAR imagery utilizes microwave imaging, enabling all-day, all-weather image acquisition. Due to its unique imaging mechanism, the interaction between radar waves and surface objects leads to phase differences between different waves, resulting in random fluctuations in intensity in the received image, thus generating speckle noise. Speckle noise is multiplicative noise, proportional to the signal amplitude, and therefore exhibits different noise intensities in different areas of the image. This can mask image texture and details, blurring the image and affecting the accuracy of change detection.

[0003] However, in actual SAR image change detection using related technologies, speckle noise may be mistaken for change, leading to false positives, reducing the accuracy of change detection, and making it difficult to identify real changes. This issue urgently needs to be addressed. Summary of the Invention

[0004] This application provides a method and apparatus for detecting changes in SAR images with reduced interference, in order to solve the problems in the actual detection of changes in SAR images in related technologies, such as speckle noise being mistaken for changes, confusing with real changes, resulting in false positive results, reducing the accuracy of change detection, and making it difficult to identify real changes.

[0005] The first aspect of this application provides a method for detecting changes in SAR images with reduced interference, comprising the following steps: identifying each pixel of the SAR image to be detected, and generating a patch of target size based on the center of each pixel; inputting the patch into a preset encoding model to obtain an encoded feature map with the interference reduced to the target interference level; inputting the encoded feature map into a preset decoding model to generate a cascaded feature map, and generating a change detection result of the SAR image based on the cascaded feature map.

[0006] Optionally, in one embodiment of this application, before inputting the patch into the preset coding model, the method further includes: extracting an initial feature map of the target image, converting the initial feature map from the spatial domain to the frequency domain to obtain a feature map in the frequency domain; processing the feature map in the frequency domain to obtain a processed feature map, and inputting the processed feature map into the target network model to output an intermediate feature map to generate the preset coding model.

[0007] Optionally, in one embodiment of this application, before inputting the encoded feature map into the preset decoding model to generate the cascaded feature map, the method further includes: splicing the intermediate feature maps to obtain a complete feature map; converting the complete feature map from the frequency domain back to the spatial domain, restoring the size of the complete feature map to its original size, and generating the preset decoding model.

[0008] Optionally, in one embodiment of this application, the method further includes: inputting the processed feature map into a target network model to obtain spatial and positional information of the processed feature map; and generating the preset encoding model based on the spatial and positional information.

[0009] Optionally, in one embodiment of this application, the step of stitching the intermediate feature maps to obtain a complete feature map includes: using multi-scale attention to obtain multi-scale features of each feature map in the stitching process; and generating the complete feature map in the channel dimension of the target network model based on the multi-scale features.

[0010] A second aspect of this application provides a SAR image change detection device with reduced interference, comprising: an identification module for identifying each pixel of a SAR image to be detected and generating a patch of target size based on the center of each pixel; a first processing module for inputting the patch into a preset encoding model to obtain an encoded feature map with interference reduced to the target interference level; and a detection module for inputting the encoded feature map into a preset decoding model to generate a cascaded feature map and generating a change detection result of the SAR image based on the cascaded feature map.

[0011] Optionally, in one embodiment of this application, it further includes: a first conversion module, configured to extract an initial feature map of the target image before inputting the patch into the preset coding model, and convert the initial feature map from the spatial domain to the frequency domain to obtain a feature map in the frequency domain; a second processing module, configured to process the feature map in the frequency domain to obtain a processed feature map, and input the processed feature map into the target network model to output an intermediate feature map to generate the preset coding model.

[0012] Optionally, in one embodiment of this application, it further includes: a splicing module, used to splice the intermediate feature map to obtain a complete feature map before inputting the encoded feature map into the preset decoding model to generate the cascaded feature map; and a second conversion module, used to convert the complete feature map from the frequency domain back to the spatial domain and restore the size of the complete feature map to its original size to generate the preset decoding model.

[0013] Optionally, in one embodiment of this application, it further includes: an acquisition module, configured to input the processed feature map into a target network model to obtain spatial information and location information of the processed feature map; and a generation module, configured to generate the preset encoding model based on the spatial information and the location information.

[0014] Optionally, in one embodiment of this application, the stitching module includes: an acquisition unit, configured to acquire multi-scale features of each feature map during the stitching process using multi-scale attention; and a generation unit, configured to generate the complete feature map in the channel dimension of the target network model based on the multi-scale features.

[0015] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the SAR image change detection method with reduced interference as described in the above embodiments.

[0016] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for detecting changes in SAR images with reduced interference.

[0017] A fifth aspect of this application provides a computer program product, including a computer program that, when executed, is used to implement the above-described method for detecting changes in SAR images with reduced interference.

[0018] This application embodiment utilizes a specific encoding model to transform the feature map of the SAR image to be detected from the spatial domain to the frequency domain. This allows for the separation of high-frequency noise from the feature map and the encoding of spatial and location information. A decoding model then bridges multi-scale features and transforms the feature map back from the frequency domain to the spatial domain, thereby completing SAR image change detection. This achieves the goal of reducing the impact of noise on change detection by separating high-frequency noise and extracting higher-dimensional information using multi-layer convolution. This information is then input into the ConvMixer module to obtain mixed spatial and location information. Furthermore, multi-scale attention gates are used to stitch together low-level and high-level features, and encoder features are adaptively selected according to different resolutions, thus improving the accuracy of SAR image change detection. This solves the problem in related technologies where speckle noise may be mistakenly identified as change in actual SAR image change detection, leading to false positives, reduced accuracy, and difficulty in identifying genuine changes.

[0019] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0020] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0021] Figure 1 This is a flowchart of a SAR image change detection method for reducing interference according to an embodiment of this application;

[0022] Figure 2 This is a schematic diagram illustrating the principle of an encoding model according to an embodiment of this application;

[0023] Figure 3 This is a schematic diagram illustrating the principle of a decoding model according to an embodiment of this application;

[0024] Figure 4 This is a schematic diagram of the structure of the SAR image change detection device for reducing interference according to the embodiments of this application;

[0025] Figure 5 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application.

[0026] Figure label:

[0027] 10-SAR image change detection device with reduced interference: 100-Identification module, 200-First processing module and 300-Detection module; 501-Memory, 502-Processor and 503-Communication interface. Detailed Implementation

[0028] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0029] The following describes a method and apparatus for detecting SAR image changes with reduced interference, based on embodiments of the present application, with reference to the accompanying drawings. Addressing the problem mentioned in the background art where speckle noise can be mistakenly identified as change in actual SAR image change detection, leading to false positives, reduced accuracy of change detection, and difficulty in identifying genuine changes, this application provides a method for detecting SAR image changes with reduced interference. In this method, a certain encoding model can be used to transform the feature map of the SAR image to be detected from the spatial domain to the frequency domain. This allows for the separation of high-frequency noise from the feature map and the encoding of spatial and location information. Then, a decoding model can be used to bridge multi-scale features and transform the feature map back from the frequency domain to the spatial domain, thereby completing the SAR image change detection. This achieves the goal of reducing the impact of noise on change detection by separating high-frequency noise and extracting higher-dimensional information using multi-layer convolution. The information is then input into a ConvMixer module to obtain mixed spatial and location information. Furthermore, multi-scale attention gates are used to stitch together low-level and high-level features, and encoder features are adaptively selected according to different resolutions, thereby improving the accuracy of SAR image change detection. This solves the problems in related technologies, such as speckle noise being mistaken for actual changes in SAR images, leading to false positives, reduced accuracy of change detection, and difficulty in identifying real changes.

[0030] Specifically, Figure 1 This is a flowchart of a SAR image change detection method for reducing interference, provided in an embodiment of this application.

[0031] like Figure 1 As shown, the SAR image change detection method with reduced interference includes the following steps:

[0032] In step S101, each pixel of the SAR image to be detected is identified, and a patch of the target size is generated based on the center of each pixel.

[0033] As will be understood by those skilled in the art, SAR imagery here refers to Synthetic Aperture Radar (SAR) imagery, an image obtained using radar signals for remote sensing imaging. A pixel is the basic unit of image display and the smallest element that makes up a digital image. Imagine an image as a jigsaw puzzle composed of many small squares; these small squares are pixels. For example, a simple black and white photograph appears as a complete portrait or landscape from a distance, but if magnified sufficiently, you will see individual small squares—these are pixels.

[0034] In some embodiments, when identifying changes in SAR images, at least a reference SAR image and a comparison SAR image are required to detect the changes in the SAR. The SAR images to be detected here can be understood as at least two SAR images for which image change detection is required.

[0035] Identifying each pixel in a SAR image to be detected can be understood as follows: a SAR image is composed of many small squares (pixels), and each small square is marked or located. These pixels contain various information about the observed scene, such as the reflectivity of ground features.

[0036] A patch can be understood as a small image block. Generating a patch of the target size is based on the center of each pixel; that is, a specific area of ​​a certain size is taken from the center of each pixel as a small image block. If the center of the current pixel is a coordinate, then the patch is an M×N square region of size 5 pixels, centered on the current pixel and encompassing a certain range of pixels in all directions. The pixels within this small square region constitute the content of the patch.

[0037] For example, the target size is defined as M×N patches of size 5 pixels each, which are M×N 5×5 square regions or other shapes with a side length of 5. Here, "size 5" can be understood as the number of pixels in the patch in both the horizontal and vertical directions. For instance, a small image patch of size 5 is formed by selecting two pixels above, below, to the left, and to the right of a given pixel (which, together with the center pixel itself, forms a 5×5 region).

[0038] It should be noted that the specific target size of the patch can be set or adjusted by those skilled in the art according to the actual situation. This is only an example and no specific limitation is made.

[0039] The embodiments of this application can generate a patch of target size centered on each pixel of the SAR image to be detected, thereby making it easier to analyze the features of pixels in each small area and facilitating target detection or feature extraction in subsequent processes.

[0040] Step S102: Input the patch into the preset encoding model to obtain the encoded feature map with the interference reduced to the target interference level.

[0041] It is understood that the preset coding model here refers to a pre-set coding model that can transform the spatial domain features of the SAR image to be detected to the frequency domain. This allows for the separation of the high-frequency components (i.e., high-frequency noise) of the SAR image, reducing noise interference to target interference and thus minimizing the impact of noise on the detection of changes in the SAR image. The encoded feature map can be understood here as the feature map obtained after processing the SAR image to be detected by inputting it into a specific coding model.

[0042] Here, target interference can be understood as the level of interference achieved after the noise interference in the SAR image to be detected has decreased. Feature map here refers to an important concept in Convolutional Neural Networks (CNNs). When processing the SAR image to be detected, convolutional layers slide convolution kernels across the input SAR image (here, a pixel patch of size , two channels) to perform convolution operations. The result of each convolution operation constitutes a new matrix, which is the feature map.

[0043] The following section will further explain the construction process of the preset encoding model in the embodiments of this application.

[0044] Optionally, in one embodiment of this application, before inputting the patch into the preset coding model, the method further includes: extracting an initial feature map of the target image, converting the initial feature map from the spatial domain to the frequency domain to obtain a feature map in the frequency domain; processing the feature map in the frequency domain to obtain a processed feature map, and inputting the processed feature map into the target network model to output an intermediate feature map to generate the preset coding model.

[0045] It is understandable that the target image here can be understood as some SAR images used when constructing a certain coding model. It can be the SAR image to be detected, or some other SAR images that can be used for change detection.

[0046] Figure 2 This is a schematic diagram illustrating the principle of an encoding model according to an embodiment of this application. Figure 2 As shown, in the actual execution process, when constructing a certain encoding model, the embodiments of this application can first extract the initial feature map of the target image. Here, the initial feature map can be understood as the feature map of the target image directly extracted from the initial stage of a neural network such as a convolutional layer without further processing.

[0047] For example, inputting a pixel patch of the target image (image size H×W, two channels), performing convolution, batch normalization, and ReLU activation on the image, repeating all the above processes twice, and then using max pooling for downsampling, yielding... The feature map of size, i.e., the initial feature map.

[0048] After obtaining the initial feature map, the embodiments of this application can transform the initial feature map from the spatial domain to the frequency domain. For example, the discrete wavelet transform can be used to transform the initial feature map from the spatial domain to the frequency domain, thereby obtaining the feature map in the frequency domain.

[0049] After obtaining the feature map in the frequency domain, this embodiment further processes the feature map in the frequency domain, that is, it divides the frequency domain features of the feature map into high-frequency features and low-frequency features, and analyzes the high-frequency features to separate high-frequency noise. It should be noted that in this embodiment, the size of the feature map after conversion to the frequency domain and separation of high-frequency noise remains unchanged, just like the size of the initial feature map.

[0050] Next, the feature map from which high-frequency noise has been separated is input into a downsampling module consisting of convolution, batch normalization, ReLU activation function, and max pooling layer for processing, and this process is repeated twice. The sizes of the two output feature maps can be expressed as follows:

[0051] After this processing, the feature map can be input into the target network model and output an intermediate feature map in this embodiment. Here, the target network model can be understood as a network model capable of extracting features and fusing information from the feature map, such as the ConMixer module.

[0052] Therefore, the embodiments of this application can combine the above processes to extract higher-dimensional information through multi-layer convolution, reduce noise interference when performing change detection on target images, and thus generate a certain coding model.

[0053] Optionally, in one embodiment of this application, the method further includes: inputting the processed feature map into the target network model to obtain the spatial and positional information of the processed feature map; and generating a preset encoding model based on the spatial and positional information.

[0054] Based on the descriptions of other embodiments, it is understood that this application inputs the processed feature map into the target network model to obtain an intermediate feature map and outputs it, so that the output intermediate feature map can be input into a certain decoding model for processing.

[0055] For example, this application may, but is not limited to, input the processed feature map into the ConMixer module to obtain the spatial and positional information of the feature map, reduce the impact of noise, and complete the construction of the entire coding model.

[0056] Step S103: Input the encoded feature map into the preset decoding model to generate a cascaded feature map, and generate the change detection result of the SAR image based on the cascaded feature map.

[0057] It is understandable that the preset decoding model here refers to a certain pre-built decoding model, which can use multi-scale attention gates to concatenate low-level and high-level features and adaptively select encoder features according to different resolutions, thereby improving the accuracy of change detection.

[0058] As one possible approach, this application can input the encoded feature map into a certain decoding model to generate a cascaded feature map of the SRA image to be detected, and then obtain the change detection result of the SRA image to be detected based on the cascaded feature map. The change detection result refers to the difference result between two or more SRA images to be detected.

[0059] Cascaded feature maps refer to the result of sequentially connecting feature maps obtained from multiple different levels or stages of processing. In the Convolutional Neural Network (CNN) architecture of deep learning, as the network depth increases, each layer generates a feature map, which contains information from different levels extracted progressively from the original image.

[0060] For example, in a simple three-layer CNN, the first layer might extract basic low-level information such as edges and lines of the image to obtain features. Figure 1 The second layer further extracts some simple shape information to obtain features. Figure 2 The third layer may extract more complex information about object components or the object as a whole, thus obtaining features. Figure 3 By concatenating these three feature maps together, a cascaded feature map is formed.

[0061] Cascaded feature maps can effectively fuse information from different levels. Low-level feature maps contain basic texture and edge information of the image, while high-level feature maps contain more abstract overall object or semantic information. Through cascading, the network can utilize this information simultaneously, enabling the SAR image to be detected to perform change detection based on a more comprehensive feature representation, resulting in more accurate change detection results.

[0062] It should be noted that the encoding model and decoding model in the embodiments of this application are not necessarily separate entities. They can also be integrated into a network to achieve change detection of the SAR image to be detected.

[0063] Optionally, in one embodiment of this application, before inputting the encoded feature map into the preset decoding model to generate the cascaded feature map, the method further includes: splicing intermediate feature maps to obtain a complete feature map; converting the complete feature map from the frequency domain back to the spatial domain and restoring the size of the complete feature map to its original size to generate the preset decoding model.

[0064] Based on the descriptions of other embodiments, it will be understood that this application can input the encoded feature map into a certain decoding model to generate a cascaded feature map.

[0065] Prior to this, this application requires concatenating intermediate feature maps from the final output of the target network model in a certain encoding model to obtain a complete feature map; then, convolution, batch normalization, and ReLU activation function operations are performed on the complete feature map, repeated twice; finally, discrete wavelet inverse transform is performed on the complete feature map to convert it from the frequency domain back to the spatial domain and restore its size to its original size, thus completing the image decoding and generating a certain decoding model. For example... Figure 3 As shown, Figure 3 This is a schematic diagram illustrating the principle of a decoding model according to one embodiment of this application.

[0066] It should be noted that the intermediate feature maps in this embodiment and the encoded feature maps in the previous embodiments are both obtained by the output of a certain encoding model. Here, the intermediate feature map refers to the feature map output during the training (construction) of the encoding model, while the encoded feature map refers to the feature map output when the encoding model is actually applied.

[0067] Optionally, in one embodiment of this application, splicing intermediate feature maps to obtain a complete feature map includes: using multi-scale attention to acquire multi-scale features of each feature map in the splicing process; and generating a complete feature map in the channel dimension of the target network model based on the multi-scale features.

[0068] In actual implementation, this embodiment of the application mainly utilizes multi-scale attention to acquire multi-scale features of each feature map during the stitching process when stitching intermediate feature maps. Then, skip connections are used in the channel dimension of the target network model to bridge the multi-scale information of the feature maps in the encoding model with the decoded information to achieve the stitching of intermediate feature maps, thereby realizing multi-level utilization of multi-scale features. The specific process can be represented as follows:

[0069] (1) Upsample, convolution, batch normalize, and perform ReLU activation function operations on the intermediate feature map output by the target network model to obtain a size of Feature map;

[0070] (2) In the coding model The feature map of the same size is input into the multi-scale attention gate, and after obtaining the multi-scale information of the same size, it is spliced ​​with the feature map generated by (1) in the channel dimension.

[0071] (3) Perform convolution, batch normalization, and ReLU activation on the initially stitched image. Repeat this process twice. Then perform upsampling, convolution, batch normalization, and ReLU activation to obtain an image of size [size missing]. Feature map;

[0072] (4) In the coding model The feature map is input into the multi-scale attention gate, and after obtaining multi-scale information of the same size, it is spliced ​​with the feature map generated by (3) in the channel dimension to obtain the complete feature map.

[0073] It should be noted that the complete feature map and the concatenated feature map in the embodiments of this application both refer to the concatenated feature map. Here, the complete feature map refers to the concatenated feature map during the training (construction) of the decoding model, while the concatenated feature map refers to the concatenated feature map when the decoding model is actually applied.

[0074] According to the SAR image change detection method with reduced interference proposed in this application, a certain coding model can be used to transform the feature map of the SAR image to be detected from the spatial domain to the frequency domain. This allows for the separation of high-frequency noise from the feature map and the encoding of spatial and location information. Then, a decoding model can be used to bridge multi-scale features and transform the feature map back from the frequency domain to the spatial domain, thereby completing the SAR image change detection. This achieves the goal of reducing the impact of noise on change detection by separating high-frequency noise and extracting higher-dimensional information using multi-layer convolution. This information is then input into the ConvMixer module to obtain mixed spatial and location information. Furthermore, multi-scale attention gates are used to stitch together low-level and high-level features, and encoder features are adaptively selected according to different resolutions, thereby improving the accuracy of SAR image change detection. This solves the problems in related technologies where speckle noise may be mistaken for change in actual SAR image change detection, leading to false positives, reduced accuracy, and difficulty in identifying real changes.

[0075] Next, referring to the accompanying drawings, a SAR image change detection device for reducing interference is described according to an embodiment of this application.

[0076] Figure 4 This is a schematic diagram of the structure of the SAR image change detection device for reducing interference according to an embodiment of this application.

[0077] like Figure 4As shown, the SAR image change detection device 10 with reduced interference includes: an identification module 100, a processing module 200 and a detection module 300.

[0078] The identification module 100 is used to identify each pixel of the SAR image to be detected and generate a patch of target size based on the center of each pixel.

[0079] The first processing module 200 is used to input the patch into the preset encoding model to obtain the encoded feature map with the interference reduced to the target interference.

[0080] The detection module 300 is used to input the encoded feature map into a preset decoding model to generate a cascaded feature map, and generate the change detection result of the SAR image based on the cascaded feature map.

[0081] Optionally, in one embodiment of this application, it further includes: a first conversion module and a second processing module.

[0082] The first conversion module is used to extract the initial feature map of the target image before inputting the patch into the preset coding model, and convert the initial feature map from the spatial domain to the frequency domain to obtain the feature map in the frequency domain.

[0083] The second processing module is used to process the feature map in the frequency domain, obtain the processed feature map, and input the processed feature map into the target network model to output the intermediate feature map to generate the preset encoding model.

[0084] Optionally, in one embodiment of this application, it further includes: a splicing module and a second conversion module.

[0085] The splicing module is used to splice intermediate feature maps to obtain a complete feature map before inputting the encoded feature map into the preset decoding model to generate the cascaded feature map.

[0086] The second conversion module is used to convert the complete feature map from the frequency domain back to the spatial domain, restoring the size of the complete feature map to its original size, so as to generate a preset decoding model.

[0087] Optionally, in one embodiment of this application, it further includes an acquisition module and a generation module.

[0088] The acquisition module is used to input the processed feature map into the target network model in order to obtain the spatial and positional information of the processed feature map.

[0089] The generation module is used to generate a preset encoding model based on spatial and location information.

[0090] Optionally, in one embodiment of this application, the splicing module includes: an acquisition unit and a generation unit.

[0091] The acquisition unit is used to acquire multi-scale features of each feature map in the stitching process using multi-scale attention.

[0092] The generation unit is used to generate a complete feature map in the channel dimension of the target network model based on multi-scale features.

[0093] It should be noted that the foregoing explanation of the SAR image change detection method for reducing interference also applies to the SAR image change detection device for reducing interference in this embodiment, and will not be repeated here.

[0094] The SAR image change detection device for reducing interference proposed in this application can utilize a certain encoding model to transform the feature map of the SAR image to be detected from the spatial domain to the frequency domain. This allows for the separation of high-frequency noise from the feature map and the encoding of spatial and location information. Then, a decoding model can bridge multi-scale features and transform the feature map back from the frequency domain to the spatial domain, thereby completing the SAR image change detection. This achieves the goal of reducing the impact of noise on change detection by separating high-frequency noise and extracting higher-dimensional information using multi-layer convolution. The information is then input into a ConvMixer module to obtain mixed spatial and location information. Furthermore, multi-scale attention gates are used to stitch together low-level and high-level features, and encoder features are adaptively selected according to different resolutions, thereby improving the accuracy of SAR image change detection. This solves the problems in related technologies where speckle noise may be mistakenly identified as change in actual SAR image change detection, leading to false positives, reduced accuracy, and difficulty in identifying real changes.

[0095] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0096] The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0097] When the processor 502 executes the program, it implements the SAR image change detection method with reduced interference provided in the above embodiments.

[0098] Furthermore, electronic devices also include:

[0099] Communication interface 503 is used for communication between memory 501 and processor 502.

[0100] The memory 501 is used to store computer programs that can run on the processor 502.

[0101] The memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0102] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0103] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0104] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0105] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for detecting changes in SAR images with reduced interference.

[0106] This application also provides a computer program product, including a computer program that can run computer instructions. When the computer instructions are executed by a processor, they implement the SAR image change detection method with reduced interference provided in this application.

[0107] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0108] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0109] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0110] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0111] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0112] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0113] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0114] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for change detection in SAR images with reduced interference, characterized in that, The method comprises the following steps: identifying each pixel point of a SAR image to be detected, and generating a patch of a target size based on the center of each pixel point; inputting the patch into a preset encoding model to obtain an encoded feature map with a reduced interference degree to a target interference degree; inputting the encoded feature map into a preset decoding model to generate a cascaded feature map, and generating a change detection result of the SAR image based on the cascaded feature map; wherein the step of inputting the patch into the preset encoding model to obtain the encoded feature map with the reduced interference degree to the target interference degree comprises: converting the feature map from a spatial domain to a frequency domain by using the preset encoding model to separate high-frequency noise; and inputting the high-frequency noise into a ConvMixer module to obtain spatial information and position information of the feature map; wherein the step of inputting the encoded feature map into the preset decoding model to generate the cascaded feature map and generating the change detection result of the SAR image based on the cascaded feature map comprises: splicing bottom layer features and high layer features by using a multi-scale attention gate in the preset decoding model, and adaptively selecting encoder features according to different resolutions.

2. The method of claim 1, wherein, Before the patch is input into the preset encoding model, the method further comprises: extracting an initial feature map of a target image, converting the initial feature map from a spatial domain to a frequency domain to obtain a feature map in the frequency domain; processing the feature map in the frequency domain to obtain a processed feature map, and inputting the processed feature map into a target network model to output an intermediate feature map to generate the preset encoding model.

3. The method of claim 2, wherein, Before the encoded feature map is input into the preset decoding model to generate the cascaded feature map, the method further comprises: splicing the intermediate feature map to obtain a complete feature map; converting the complete feature map from the frequency domain back to the spatial domain, restoring the size of the complete feature map to the original size to generate the preset decoding model.

4. The method of claim 2, wherein, The method further comprises: inputting the processed feature map into the target network model to obtain spatial information and position information of the processed feature map; generating the preset encoding model based on the spatial information and the position information.

5. The method of claim 3, wherein, The step of splicing the intermediate feature map to obtain a complete feature map comprises: obtaining multi-scale features of each feature map in the splicing process by using a multi-scale attention; generating the complete feature map in the channel dimension of the target network model based on the multi-scale features.

6. An apparatus for change detection of SAR images with reduced interference, characterized in that The method comprises: a generation module configured to identify each pixel point of a SAR image to be detected, and generate a patch of a target size based on the center of each pixel point; a first processing module configured to input the patch into a preset encoding model to obtain an encoded feature map with a reduced interference degree to a target interference degree; a detection module configured to input the encoded feature map into a preset decoding model to generate a cascaded feature map, and generate a change detection result of the SAR image based on the cascaded feature map. The first processing module comprises: converting the feature map from a spatial domain to a frequency domain by using the preset encoding model to separate high-frequency noise; inputting the high-frequency noise into a ConvMixer module to obtain spatial information and position information of the feature map. The detection module comprises: splicing bottom layer features and high layer features by using a multi-scale attention gate in the preset decoding model, and adaptively selecting encoder features according to different resolutions.

7. The apparatus of claim 6, wherein, Further comprising: A conversion module is configured to extract an initial feature map of the target image before inputting the patch into the preset encoding model, convert the initial feature map from a spatial domain to a frequency domain, and obtain a feature map in the frequency domain; A second processing module is configured to process the feature map in the frequency domain to obtain a processed feature map, input the processed feature map into a target network model, output an intermediate feature map, and generate the preset encoding model.

8. An electronic device, comprising: Further comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the SAR image change detection method with reduced interference according to any one of claims 1-5.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the SAR image change detection method with reduced interference according to any one of claims 1-5.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed to implement the SAR image change detection method with reduced interference according to any one of claims 1-5.

Images