Abnormity detection method and system based on synthetic aperture radar

Abstract

The invention relates to an anomaly detection method and system based on a synthetic aperture radar, and the method comprises the steps: collecting an SAR image of the synthetic aperture radar; the method comprises the following steps: denoising an SAR image to obtain a denoised image; performing image reconstruction on the de-noised image through a self-supervised model based on a generative adversarial network to obtain a reconstructed image of a highlighted abnormal target; generating an anomaly graph representing the pixel difference between the denoised image and the reconstructed image through a change detection algorithm based on a covariance matrix; and determining an abnormal target in the abnormal graph through a constant false alarm detection algorithm. According to the method and the device, the high-quality de-noised SAR image is firstly obtained, then the image is reconstructed to highlight the abnormal target in the image, and finally, all anomalies are further filtered on the basis of detecting the abnormal points in the image, so that the false alarm rate is reduced while the abnormal points in the image can be accurately detected, and the accuracy of the detection result is improved. The problem of how to improve the anomaly detection effect of the synthetic aperture radar image is solved.

Description

Technical Field

[0001] The present application relates to the field of aerospace surveying and mapping technology, and in particular to an anomaly detection method and system based on synthetic aperture radar. Background Art

[0002] Synthetic Aperture Radar (SAR) is designed to provide images of the Earth's surface in radar frequency bands. Unlike optical imaging, SAR can work day and night by scanning an area, with the radar actively transmitting and receiving pulses, that is, SAR can monitor changes and anomalies on the Earth's surface in real time. Anomaly detection based on SAR imaging aims to discover abnormal patterns hidden in multi-dimensional radar signals and images. Such anomalies may be man-made changes in a specific location or natural processes affecting a specific area. For example, these anomalies can characterize a variety of potential applications: oil slick detection, ship wakes, dam anomalies, ground vehicles, aircraft, etc.

[0003] Deep learning technology is usually used to perform anomaly detection and obtain anomaly detection results. In addition, there are also statistical machine learning methods such as SVM that analyze the numerical features of SAR images to detect anomalies. However, deep learning is limited by training data and does not have good generalization capabilities. It usually only has good effects in specific scenarios; and statistical learning has defects in modeling the spatial correlation of SAR images, and the detection effect is also poor.

[0004] Currently, no effective solution has been proposed for how to improve the anomaly detection effect of synthetic aperture radar images in related technologies. Summary of the invention

[0005] The embodiments of the present application provide a method and system for anomaly detection based on synthetic aperture radar, so as to at least solve the problem of how to improve the anomaly detection effect of synthetic aperture radar images in the related art.

[0006] In a first aspect, an embodiment of the present application provides an anomaly detection method based on a synthetic aperture radar image, the method comprising:

[0007] Collect synthetic aperture radar imaging data to obtain SAR images;

[0008] Denoising the SAR image using a trained denoising model to obtain a denoised image;

[0009] Reconstructing the denoised image through a self-supervised model based on a generative adversarial network to obtain a reconstructed image that highlights the abnormal target;

[0010] generating an anomaly map characterizing pixel differences between the denoised image and the reconstructed image by a change detection algorithm based on a covariance matrix;

[0011] The abnormal target in the abnormal image is confirmed by a constant false alarm detection algorithm.

[0012] In some embodiments, before reconstructing the denoised image by a self-supervised model based on a generative adversarial network to obtain a reconstructed image highlighting abnormal targets, the method includes:

[0013] Constructing a self-supervised model based on a generative adversarial network, wherein the self-supervised model includes a generator and a recognizer;

[0014] The self-supervisory model is trained so that the reconstruction error of normal pixels in the reconstructed image obtained by the self-supervisory model is reduced as much as possible, and the distribution characteristics of the reconstructed image are kept consistent with the denoised image as much as possible.

[0015] In some embodiments, generating an anomaly map characterizing pixel differences between the denoised image and the reconstructed image by a change detection algorithm based on a covariance matrix includes:

[0016] Calculating the maximum likelihood estimate of the average vector of pixels in the sliding window in the denoised image and the reconstructed image respectively;

[0017] Calculating the covariance matrices of the denoised image and the reconstructed image respectively based on the maximum likelihood estimation of the average vector of the pixels in the sliding window;

[0018] Based on the covariance matrix of the denoised image and the reconstructed image, an anomaly map characterizing pixel differences between the denoised image and the reconstructed image is generated.

[0019] In some embodiments, respectively calculating the maximum likelihood estimate of the average vector of pixels within the sliding window in the denoised image and the reconstructed image includes:

[0020] By averaging the vector sampling formula Calculate the maximum likelihood estimate of the average vector of the pixels in the sliding window in the denoised image and the maximum likelihood estimate of the average vector of the pixels in the sliding window in the reconstructed image, where β k,l represents the sliding window with the center pixel coordinate (k, l), X i,j Represents the pixel with coordinates (i, j) in the denoised image or the reconstructed image.

[0021] In some embodiments, calculating the covariance matrices of the denoised image and the reconstructed image respectively based on the maximum likelihood estimation of the average vector of pixels in the sliding window includes:

[0022] Through the covariance sampling formula The covariance matrix of the denoised image and the covariance matrix of the reconstructed image are calculated, where: represents the maximum likelihood estimate of the mean vector of pixels in the sliding window in the denoised image or the reconstructed image, β k,l represents the sliding window with the center pixel coordinate (k, l), X i,j Represents the pixel with coordinates (i, j) in the denoised image or the reconstructed image.

[0023] In some embodiments, generating an anomaly map representing pixel differences between the denoised image and the reconstructed image based on a covariance matrix of the denoised image and the reconstructed image comprises:

[0024] Generate formulas through anomaly graphs An anomaly map characterizing pixel differences between the denoised image and the reconstructed image is generated, wherein: represents the covariance matrix of the denoised image, represents the covariance matrix of the reconstructed image, and F represents the calculation of the Euclidean distance between the two covariance matrices.

[0025] In some embodiments, before denoising the SAR image using a trained denoising model, the method includes:

[0026] Get historical synthetic aperture radar SAR images;

[0027] Based on the historical SAR images, a noise sample data set for denoising model training is obtained;

[0028] A denoising model is trained based on the noise sample data set.

[0029] In some embodiments, obtaining a noise sample data set for denoising model training based on the historical SAR images includes:

[0030] Screening out historical SAR images without speckle noise from the historical SAR images;

[0031] Based on the historical SAR image without speckle noise, constructing a historical SAR image with added noise;

[0032] Based on the historical SAR images without speckle noise and the historical SAR images with added noise, a noise sample data set for denoising model training is constructed.

[0033] In some embodiments, training a denoising model based on the noise sample dataset includes:

[0034] During the training process of the denoising model, the historical SAR image with noise added in the noise sample data set is input into the denoising model, and the denoised image is output;

[0035] The difference between the denoised image and the historical SAR image without speckle noise in the noise sample data set is minimized by a preset loss function, and the network parameters of the denoising model are adjusted accordingly.

[0036] In a second aspect, an embodiment of the present application provides an anomaly detection system based on a synthetic aperture radar image, the system being used to execute the method described in any one of the first aspects above, the system comprising a data acquisition module, a data denoising module, an image reconstruction module and an anomaly detection module;

[0037] The data acquisition module is used to collect imaging data of synthetic aperture radar to obtain SAR images;

[0038] The data denoising module is used to denoise the SAR image using a trained denoising model to obtain a denoised image;

[0039] The image reconstruction module is used to reconstruct the denoised image through a self-supervised model based on a generative adversarial network to obtain a reconstructed image that highlights the abnormal target;

[0040] The anomaly detection module is used to generate an anomaly map characterizing the pixel differences between the denoised image and the reconstructed image through a change detection algorithm based on a covariance matrix; and to confirm abnormal targets in the anomaly map through a constant false alarm detection algorithm.

[0041] Compared with the related art, the embodiment of the present application provides a method and system for detecting anomalies based on synthetic aperture radar, wherein the method obtains a SAR image by collecting imaging data of a synthetic aperture radar; denoises the SAR image by a trained denoising model to obtain a denoised image; reconstructs the denoised image by a self-supervised model based on a generative adversarial network to obtain a reconstructed image highlighting abnormal targets; generates an anomaly map characterizing the pixel differences between the denoised image and the reconstructed image by a change detection algorithm based on a covariance matrix; confirms abnormal targets in the anomaly map by a constant false alarm detection algorithm, and realizes first obtaining a high-quality denoised SAR image by a denoising model, and then reconstructing the image by an adversarial network model to highlight the abnormal targets therein; finally, by a covariance detection algorithm and a constant false alarm detection algorithm, all anomalies are further filtered on the basis of detecting abnormal points in the image, thereby ensuring that the abnormal points in the image can be accurately detected, while reducing the false alarm rate, and solving the problem of how to improve the anomaly detection effect of synthetic aperture radar images. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0043] Figure 1 is a flowchart of the steps of the anomaly detection method based on synthetic aperture radar images according to an embodiment of the present application;

[0044] Figure 2 is a schematic diagram of a denoised image according to an embodiment of the present application;

[0045] Figure 3 is a schematic diagram of an abnormal diagram according to an embodiment of the present application;

[0046] Figure 4 is a schematic diagram of the process of SAR image slice traversal processing according to an embodiment of the present application;

[0047] Figure 5 is a structural block diagram of an anomaly detection system based on synthetic aperture radar images according to an embodiment of the present application;

[0048] Figure 6 It is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0050] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0051] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0052] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0053] The present application embodiment provides an anomaly detection method based on synthetic aperture radar images. Figure 1 is a flowchart of the steps of the anomaly detection method based on synthetic aperture radar images according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0054] Step S102, collecting imaging data of synthetic aperture radar to obtain a SAR image;

[0055] It should be noted that SAR (Synthetic Aperture Radar) imagery is a technology that creates detailed images of the ground or other surfaces by using radar waves. Unlike optical imaging, SAR can work in all weather conditions and day and night because it relies on microwaves instead of visible light. This means that SAR images can obtain information about the Earth's surface under cloud cover, at night, or other conditions where optical sensors cannot work. Speckle noise is a physical phenomenon that radar cannot distinguish. It is caused by the coherent sum of contributions from different basic scatterers within the same resolution unit. In other words, it is a phase difference that causes fluctuations in the complex sum, which then causes fluctuations in the observed amplitude, resulting in granular behavior in the observed image. Since speckle has a very strong impact on the quality of SAR imaging and makes the image difficult to analyze, removing speckle noise in SAR images is crucial to the effectiveness of subsequent anomaly detection models.

[0056] Step S104, denoising the SAR image using the trained denoising model to obtain a denoised image;

[0057] Before step S104, the method further includes step S103, acquiring historical SAR images of synthetic aperture radar; obtaining a noise sample data set for denoising model training based on the historical SAR images; and training the denoising model based on the noise sample data set.

[0058] Step S103 specifically includes the following steps:

[0059] Step S1031, screening out historical SAR images without speckle noise from the acquired historical SAR images; constructing historical SAR images with added noise based on the historical SAR images without speckle noise; and constructing a noise sample data set for denoising model training based on the historical SAR images without speckle noise and the historical SAR images with added noise.

[0060] It should be noted that, assuming that speckle noise follows a gamma distribution, a large number of noisy samples are generated by adding speckle noise to normal historical SAR images for preliminary training of the model to make up for the lack of SAR image samples in general, and then fine-tuning is performed on real SAR images containing speckle noise. This overcomes the limitations of common data-driven methods in modeling denoising models due to the lack of data diversity.

[0061] Step S1032, during the training process of the denoising model, the historical SAR image with noise added in the noise sample data set is input into the denoising model to output the denoised image; the difference between the denoised image and the historical SAR image without speckle noise in the noise sample data set is minimized by a preset loss function, and the network parameters of the denoising model are adjusted accordingly.

[0062] It should be noted that when training with a training set constructed from multiple SAR image sequences, the network is continuously optimized using a preset loss function to achieve the construction of a denoising model. That is, a pair of images (noise-noise pairing) is randomly selected from the training set, and after processing, it enters the network of the denoising model. The neural network model predicts the restored state of its noise-free image, calculates the loss with the original noise-free image, and back-propagates to optimize the network parameters.

[0063] Step S106, reconstructing the denoised image through a self-supervised model based on a generative adversarial network to obtain a reconstructed image that highlights the abnormal target;

[0064] It should be noted that, after the image to be detected is denoised in the previous steps, the natural noise caused by the device or environment will be effectively removed. The following goal is to use the self-supervised model to display the real abnormal target in the image more clearly to improve the subsequent detection effect. The self-supervised model proposed in this embodiment is based on GAN (Generative Adversarial Network). GAN is a self-supervised deep learning algorithm used to generate images similar to its input, and is trained by a generator and a recognizer at the same time. The purpose of the generator is to generate images that are as close to the real situation as possible, and the purpose of the recognizer is to distinguish the difference between the results of the generator and the real image. The more real the generator is through training, the more difficult it is for the recognizer to distinguish between true and false, and the stronger the ability of the recognizer to distinguish between true and false through training, the more difficult it is for the image generated by the generator to "cheat" the recognizer, and the two confront each other, and finally make the generator's generation result close to the real image.

[0065] Before step S106, the method also includes step S105, constructing a self-supervised model based on a generative adversarial network, wherein the self-supervised model includes a generator and a recognizer; training the self-supervised model so that the reconstruction error of normal pixels in the reconstructed image obtained by the self-supervised model is reduced as much as possible, and the distribution characteristics of the reconstructed image are kept as consistent as possible with the denoised image.

[0066] It should be noted that the generator in the self-supervised model constructed in this embodiment is composed of an encoder and a decoder, and the recognizer is between the encoder and the decoder, mainly used to maintain its latent spatial feature distribution characteristics. The input of the model is a denoised SAR image slice X∈R h*w*c , where h is the height of the image, w is the width of the image, and c is the number of channels of the image. After being processed by the encoder network, the latent spatial features of the image are obtained Then for this latent space feature z, a vector X'=D(z) with the same dimension as the original image slice X will be generated after the decoder network.

[0067] It should be further explained that by inputting a large number of denoised images as training sets into the self-supervised model, the self-supervised model can be continuously trained and the generator performance can be optimized. This embodiment has two training goals for the self-supervised model: one is to reduce the reconstruction error of normal pixels as much as possible, and the other is to maintain the original distribution characteristics of the entire image as much as possible. Therefore, its loss function is also divided into two parts.

[0068] The first is the reconstruction loss function, which minimizes the reconstruction difference between normal pixels as much as possible, and is defined as follows:

[0069]

[0070] Among them, h represents the image height, w represents the image width, c is the number of image channels, and X i,j,k is the pixel value of the i-th row, j-th column, and c-th channel of the image.

[0071] The second is the regularization loss function, which is used to control the distribution of image data under p(z) and is defined as follows:

[0072]

[0073] Among them, Z real is the encoded feature of the encoder on the original image, D c is a decoder, so D c (Z real ) is the feature of the real image after passing through the decoder, E Zreal-N(0,1) is the distribution estimation function, constraining Z real In the (0,1) normal distribution space. The corresponding is the encoding feature of the image generated by the generator, E is the feature of the image encoding feature generated by the generator after passing through the decoder. epsilon(X)-p(z) It is also a distribution estimation function, which constrains the images generated by the generator to be distributed in the space of epsilon(X)-p(z), where p(z) is a prior distribution.

[0074] Step S108, generating an anomaly map representing the pixel differences between the denoised image and the reconstructed image by using a change detection algorithm based on the covariance matrix;

[0075] Step S108 specifically includes the following steps:

[0076] Step S1081, respectively calculating the maximum likelihood estimate of the average vector of pixels in the sliding window in the denoised image and the reconstructed image;

[0077] Step S1081 Specifically, by the average vector sampling formula Calculate the maximum likelihood estimate of the average vector of pixels in the sliding window in the denoised image and the maximum likelihood estimate of the average vector of pixels in the sliding window in the reconstructed image, where β k,l represents the sliding window with the center pixel coordinate (k, l), X i,j Represents the pixel with coordinates (i, j) in the denoised image or the reconstructed image.

[0078] It should be noted that, in order to detect changes between images, this embodiment detects changes by testing the equality of the estimated covariance matrices of corresponding pixels in two images, and the covariance matrix is ​​estimated locally around the tested pixel. This processing method can better reduce the interference effect of imaging noise on abnormalities.

[0079] Step S1082, based on the maximum likelihood estimation of the average vector of the pixels in the sliding window, respectively calculating the covariance matrices of the denoised image and the reconstructed image;

[0080] Step S108 specifically, through the covariance sampling formula Calculate the covariance matrix of the denoised image and the covariance matrix of the reconstructed image, where represents the maximum likelihood estimate of the mean vector of pixels in the sliding window in the denoised image or the reconstructed image, β k,l represents the sliding window with the center pixel coordinate (k, l), X i,j Represents the pixel with coordinates (i, j) in the denoised image or the reconstructed image.

[0081] It should be noted that, in order to estimate the covariance matrix, this embodiment uses a sliding window β k,l , (k, l) represents the coordinates of the center pixel of the window. For a multivariate Gaussian distribution N(μ,∑), μ is the maximum likelihood estimate of the mean vector, ∑ is the covariance matrix, which can be calculated by the mean vector sampling (SMV) and the covariance sampling matrix (SCM).

[0082] Step S1083, generating an anomaly map representing the pixel differences between the denoised image and the reconstructed image based on the covariance matrix of the denoised image and the reconstructed image.

[0083] Step S1083 Specifically, the formula is generated through the abnormal graph Generate anomaly maps that characterize the pixel differences between the denoised image and the reconstructed image, where represents the covariance matrix of the denoised image, represents the covariance matrix of the reconstructed image, and F represents the calculation of the Euclidean distance between the two covariance matrices.

[0084] It should be noted that Figure 2is a schematic diagram of a denoised image according to an embodiment of the present application, wherein an abnormality is indicated by detecting a change between the denoised image X and the reconstructed image X'. Figure 3 is a schematic diagram of an abnormal diagram according to an embodiment of the present application, such as Figure 3 As shown, for the generated abnormal graph A X , which stores the abnormality degree of the corresponding pixel position of the SAR image, is a floating point number. By setting a threshold θ, when the abnormal image A X When the value of a position in is greater than θ, the value of the position remains unchanged. When it is less than θ, the value of the position is set to 0. Then its anomaly matrix (anomaly graph) can be expressed as A X ∈[0,1] h*w The closer the element in the anomaly matrix is ​​to 1, the greater the probability that it is an anomaly.

[0085] Step S110, confirming abnormal targets in the abnormal image by using a constant false alarm detection algorithm.

[0086] It should be noted that Constant False Alarm Rate (CFAR) is a signal detection technology used in radar systems. Its main purpose is to keep the false alarm probability (i.e. the probability of false detection caused by background noise or interference) constant under various backgrounds. In this way, even if the interference level in the working environment changes, the radar system can still maintain a consistent performance level and will not generate too many false alarms due to the increase in noise level.

[0087] Through the above steps in the embodiment of the present application, it is achieved that a high-quality denoised SAR image is first obtained through a denoising model, and then the image is reconstructed through an adversarial network model to highlight abnormal targets therein. Finally, through a covariance detection algorithm and a constant false alarm detection algorithm, all abnormalities are further filtered on the basis of detecting abnormal points in the image, thereby ensuring that the abnormal points in the image can be accurately detected and the false alarm rate is reduced, thereby solving the problem of how to improve the abnormality detection effect of synthetic aperture radar images.

[0088] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0089] In some of these embodiments, Figure 4 FIG. 1 is a flow chart of SAR image slice traversal processing according to an embodiment of the present application. Figure 4As shown in the figure, since SAR images are usually large, it is impossible to process a complete image at a time. Therefore, it is usually divided into several slices for processing. That is, the slices of the entire SAR image are traversed and processed, and the anomaly maps of the slices are obtained and then merged to obtain the anomaly detection result of the complete SAR image.

[0090] The embodiment of the present application provides an anomaly detection system based on synthetic aperture radar images. Figure 5 is a structural block diagram of an anomaly detection system based on synthetic aperture radar images according to an embodiment of the present application, such as Figure 5 As shown, the system includes a data acquisition module, a data denoising module, an image reconstruction module and an anomaly detection module;

[0091] A data acquisition module is used to collect imaging data of synthetic aperture radar to obtain SAR images;

[0092] A data denoising module is used to denoise the SAR image using a trained denoising model to obtain a denoised image;

[0093] An image reconstruction module is used to reconstruct the denoised image through a self-supervised model based on a generative adversarial network to obtain a reconstructed image that highlights abnormal targets;

[0094] The anomaly detection module is used to generate an anomaly map that represents the pixel differences between the denoised image and the reconstructed image through a change detection algorithm based on the covariance matrix; and to confirm abnormal targets in the anomaly map through a constant false alarm detection algorithm.

[0095] Through the data acquisition module, data denoising module, image reconstruction module and anomaly detection module in the embodiments of the present application, it is achieved that a high-quality denoised SAR image is first obtained through a denoising model, and then the image is reconstructed through an adversarial network model to highlight abnormal targets therein. Finally, through a covariance detection algorithm and a constant false alarm detection algorithm, all anomalies are further filtered on the basis of detecting abnormal points in the image, thereby ensuring that the abnormal points in the image can be accurately detected, while reducing the false alarm rate, and solving the problem of how to improve the anomaly detection effect of synthetic aperture radar images.

[0096] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0097] This embodiment further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0098] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0099] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0100] In addition, in combination with the anomaly detection method based on synthetic aperture radar images in the above embodiments, the present application embodiment may provide a storage medium for implementation. The storage medium stores a computer program; when the computer program is executed by a processor, any one of the anomaly detection methods based on synthetic aperture radar images in the above embodiments is implemented.

[0101] In one embodiment, a computer device is provided, which may be a terminal. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for detecting anomalies based on synthetic aperture radar images is implemented. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball, or touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0102] In one embodiment, Figure 6 is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application, such as Figure 6 As shown, an electronic device is provided, which may be a server, and its internal structure diagram may be as shown in Figure 6 As shown. The electronic device includes a processor, a network interface, an internal memory and a non-volatile memory connected through an internal bus, wherein the non-volatile memory stores an operating system, a computer program and a database. The processor is used to provide computing and control capabilities, the network interface is used to communicate with an external terminal through a network connection, the internal memory is used to provide an environment for the operation of the operating system and the computer program, the computer program is executed by the processor to implement an anomaly detection method based on a synthetic aperture radar image, and the database is used to store data.

[0103] Those skilled in the art will understand that Figure 6 The structure shown in the figure is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0104] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0105] Those skilled in the art should understand that the technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0106] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for detecting anomalies based on synthetic aperture radar images, characterized in that: The method comprises: Collect synthetic aperture radar imaging data to obtain SAR images; De-noising the SAR image using a trained de-noising model to obtain a de-noised image; Reconstructing the denoised image through a self-supervised model based on a generative adversarial network to obtain a reconstructed image that highlights the abnormal target; generating an anomaly map characterizing pixel differences between the denoised image and the reconstructed image by a change detection algorithm based on a covariance matrix; The abnormal target in the abnormal image is confirmed by a constant false alarm detection algorithm.

2. The method according to claim 1, characterized in that Before reconstructing the denoised image by using a self-supervised model based on a generative adversarial network to obtain a reconstructed image highlighting abnormal targets, the method includes: Constructing a self-supervised model based on a generative adversarial network, wherein the self-supervised model includes a generator and a recognizer; The self-supervisory model is trained so that the reconstruction error of normal pixels in the reconstructed image obtained by the self-supervisory model is reduced as much as possible, and the distribution characteristics of the reconstructed image are kept consistent with the denoised image as much as possible.

3. The method according to claim 1, characterized in that Generating an anomaly map characterizing the pixel difference between the denoised image and the reconstructed image by a change detection algorithm based on a covariance matrix comprises: Calculating the maximum likelihood estimate of the average vector of pixels in the sliding window in the denoised image and the reconstructed image respectively; Calculating the covariance matrices of the denoised image and the reconstructed image respectively based on the maximum likelihood estimation of the average vector of the pixels in the sliding window; Based on the covariance matrix of the denoised image and the reconstructed image, an anomaly map characterizing pixel differences between the denoised image and the reconstructed image is generated.

4. The method according to claim 3, characterized in that Calculating the maximum likelihood estimate of the average vector of pixels in the sliding window in the denoised image and the reconstructed image respectively includes: By averaging the vector sampling formula Calculate the maximum likelihood estimate of the average vector of the pixels in the sliding window in the denoised image and the maximum likelihood estimate of the average vector of the pixels in the sliding window in the reconstructed image, where β k,l represents the sliding window with the center pixel coordinate (k, l), X i,j Represents the pixel with coordinates (i, j) in the denoised image or the reconstructed image.

5. The method according to claim 3, characterized in that: Based on the maximum likelihood estimation of the average vector of pixels in the sliding window, respectively calculating the covariance matrix of the denoised image and the reconstructed image comprises: Through the covariance sampling formula The covariance matrix of the denoised image and the covariance matrix of the reconstructed image are calculated, where: represents the maximum likelihood estimate of the mean vector of pixels in the sliding window in the denoised image or the reconstructed image, β k,l represents the sliding window with the center pixel coordinate (k, l), X i,j Represents the pixel with coordinates (i, j) in the denoised image or the reconstructed image.

6. The method according to claim 3, characterized in that: Generating an anomaly map representing pixel differences between the denoised image and the reconstructed image based on a covariance matrix of the denoised image and the reconstructed image comprises: Generate formulas through anomaly graphs An anomaly map characterizing pixel differences between the denoised image and the reconstructed image is generated, wherein: represents the covariance matrix of the denoised image, represents the covariance matrix of the reconstructed image, and F represents the calculation of the Euclidean distance between the two covariance matrices.

7. The method according to claim 1, characterized in that Before denoising the SAR image using the trained denoising model, the method comprises: Get historical synthetic aperture radar SAR images; Based on the historical SAR images, a noise sample data set for denoising model training is obtained; A denoising model is trained based on the noise sample data set.

8. The method according to claim 7, characterized in that Based on the historical SAR images, a noise sample data set for denoising model training is obtained, including: Screening out historical SAR images without speckle noise from the historical SAR images; Based on the historical SAR image without speckle noise, constructing a historical SAR image with added noise; Based on the historical SAR images without speckle noise and the historical SAR images with added noise, a noise sample data set for denoising model training is constructed.

9. The method according to claim 8, characterized in that Based on the noise sample data set, training the denoising model includes: During the training process of the denoising model, the historical SAR image with noise added in the noise sample data set is input into the denoising model, and the denoised image is output; The difference between the denoised image and the historical SAR image without speckle noise in the noise sample data set is minimized by a preset loss function, and the network parameters of the denoising model are adjusted accordingly.

10. An anomaly detection system based on synthetic aperture radar images, characterized in that: The system is used to perform the method described in any one of claims 1 to 9, and the system includes a data acquisition module, a data denoising module, an image reconstruction module and an anomaly detection module; The data acquisition module is used to collect imaging data of synthetic aperture radar to obtain SAR images; The data denoising module is used to denoise the SAR image using a trained denoising model to obtain a denoised image; The image reconstruction module is used to reconstruct the denoised image through a self-supervised model based on a generative adversarial network to obtain a reconstructed image that highlights the abnormal target; The anomaly detection module is used to generate an anomaly map characterizing the pixel differences between the denoised image and the reconstructed image through a change detection algorithm based on a covariance matrix; and to confirm abnormal targets in the anomaly map through a constant false alarm detection algorithm.

Images