A change detection method with adaptive detection threshold

Abstract

The application provides a change detection method for adaptive threshold, and relates to the technical field of remote sensing image analysis, and the method comprises the following steps: performing cooperative superpixel segmentation on remote sensing images before and after change to obtain superpixel segmentation results of the remote sensing images before and after change; calculating affinity matrices of the remote sensing images before and after change; selecting superpixel-level training samples through the superpixel segmentation results and the affinity matrices of the remote sensing images before and after change; training a Copula function through the superpixel-level training samples, and generating an adaptive change detection threshold by using the trained Copula function; calculating test statistics of the remote sensing images before and after change through the trained Copula function, and performing detection and judgment on the test statistics by using the adaptive change detection threshold to obtain a change detection result. The application adopting the above scheme is suitable for change detection of remote sensing images in various remote sensing scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image analysis, and particularly relates to a change detection method with adaptive detection threshold. BACKGROUND

[0002] Change detection technology of multi-temporal space-based remote sensing images has an important practical position in remote sensing applications, and is widely used in the needs of remote sensing tasks such as environmental monitoring, urban planning and development, and real-time disaster assessment. Traditional change detection methods often face the challenge of how to robustly detect change signals in various remote sensing scenes. These methods usually rely on fixed or simple statistical characteristics to determine the detection threshold, and apply the detection threshold to the image test statistics to identify the change region. However, in actual applications, due to the complexity and variability of remote sensing images, these methods often cannot effectively adapt to the change patterns in different remote sensing scenes, which may lead to excessive false negatives or false positives, affecting the accuracy and robustness of the detection. Therefore, how to improve the existing methods and improve the adaptability in various conditions has become an important problem in practice. SUMMARY

[0003] The present application aims to at least partially solve one of the technical problems in the related art.

[0004] To this end, the first object of the present application is to provide a change detection method with adaptive detection threshold, which solves the technical problem that the existing methods cannot effectively adapt to the change patterns in different remote sensing scenes, leading to excessive false negatives or false positives, affecting the detection accuracy and robustness, and can flexibly cope with various remote sensing scenes, improving the performance of change detection, and providing a more reliable theoretical basis and practical guidance for remote sensing image analysis.

[0005] The second object of the present application is to provide a computer device.

[0006] To achieve the above object, the first aspect of the present application provides a change detection method with adaptive detection threshold, comprising: performing collaborative superpixel segmentation on remote sensing images before and after the change to obtain superpixel segmentation results of the remote sensing images before and after the change; calculating affinity matrices of the remote sensing images before and after the change; selecting superpixel-level training samples through the superpixel segmentation results of the remote sensing images before and after the change and the affinity matrices; training a Copula function through the superpixel-level training samples, and generating an adaptive change detection threshold using the trained Copula function; calculating test statistics of the remote sensing images before and after the change through the trained Copula function, and detecting the test statistics using the adaptive change detection threshold to obtain a change detection result.

[0007] The change detection method of adaptive detection threshold of the embodiment of the application introduces an adaptive threshold setting mechanism based on Copula theory, dynamically adjusts the threshold by analyzing the marginal distribution of multi-temporal remote sensing images and their joint distribution, to adapt to the change characteristics in different scenes. Specifically, first, the multi-temporal remote sensing images are preprocessed, the features are extracted, and the marginal distribution model is established, then the Copula function is used to model the dependence relationship between the images, and the adaptive change detection threshold is generated by combining the statistical learning method. This adaptive mechanism can effectively reduce false positives and false negatives, improve the accuracy of change detection, and remain robust in various complex environments. The application not only improves the performance of change detection, but also provides a more reliable theoretical basis and practical guidance for the analysis of remote sensing images.

[0008] Optionally, in an embodiment of the application, the collaborative superpixel segmentation of the remote sensing image is realized by the SLIC simple linear iterative clustering algorithm.

[0009] Before the collaborative superpixel segmentation of the remote sensing images before and after the change, further comprising: pre-processing the remote sensing images, wherein the pre-processing process comprises:

[0010] Judging whether the channel number of the remote sensing image is greater than 1, if greater than 1, performing channel dimension reduction processing on the remote sensing image.

[0011] Optionally, in an embodiment of the application, the superpixel segmentation result is a segmented superpixel map, and the superpixel-level training samples are selected by the superpixel segmentation results of the remote sensing images before and after the change and the affinity matrix, comprising:

[0012] Calculating the difference between the affinity matrices of each remote sensing image to obtain a change probability map;

[0013] Performing intersection operation on the segmented superpixel map and the change probability map to obtain a change probability set of the pixel points in each pair of superpixels;

[0014] Averaging the elements in the change probability set of the pixel points to obtain the change probability of each pair of superpixels;

[0015] According to the priority from small to large of the change probability, selecting a preset proportion of superpixels from all superpixels as training superpixels, and taking the selected training superpixels as superpixel-level training samples.

[0016] Optionally, in an embodiment of the application, the superpixel-level training samples include training superpixels of the remote sensing images before and after the change, and the Copula function is trained by the superpixel-level training samples, comprising:

[0017] Averaging the pixel intensity inside each training superpixel to obtain a superpixel feature value of each superpixel;

[0018] combining the superpixel feature values of all the training superpixels corresponding to each remote sensing image into a training superpixel feature value set of the remote sensing image;

[0019] performing a probability integral transformation on the training superpixel feature value set of each remote sensing image to transform the training superpixel feature values into probability distribution function value features;

[0020] inputting the probability distribution function value features of the remote sensing images before and after the change into a Copula function for training and fitting, extracting the correlation between the probability distribution function value features, and obtaining the optimal parameters of the Copula function.

[0021] Optionally, in an embodiment of the present application, the remote sensing images before and after the change are X and Y are a pair of registered remote sensing images of the same region obtained at different time points T1 and T2, the training superpixel feature sets corresponding to X and Y are M and N, and the probability distribution function value features are:

[0022] U = F M (M)

[0023] V = F N (N)

[0024] wherein F M is the probability distribution function of M, and F N is the probability distribution function of N;

[0025] The optimal parameters of the Copula function are:

[0026]

[0027] wherein θopt represents the optimal parameters found, θ represents the internal parameter of the current Copula function, the expression of is not explicitly written, and the Copula density thereof is expressed as Copula'(U, V).

[0028] Optionally, in an embodiment of the present application, an adaptive change detection threshold is generated by using the trained Copula function, including:

[0029] ​The change detection problem is transformed into a binary hypothesis testing problem using a trained Copula function. The binary hypotheses include a change hypothesis and a no-change hypothesis. Let H1 be the change hypothesis and H0 be the no-change hypothesis. The Copula function is the joint distribution function of the value features U and V under the no-change and change hypotheses, respectively: Copula(U,V|H0) and Copula(U,V|H1). The Copula function is the probability density function of the value features U and V under the no-change and change hypotheses, respectively: fi. c (U,V|H0), f c (U,V|H1), under the assumption of variation, the remote sensing image regions corresponding to different times follow independent distributions, and the probability density function of the Copula function of the remote sensing image regions corresponding to different times is 1, expressed as: f c Given (U,V|H1)=1, and assuming no changes, the distribution function of the remote sensing images at different times is f. c (U,V|H0);

[0030] According to the Neyman-Pearson criterion, given a fixed false alarm rate, the likelihood ratio test is used to obtain the test statistic.

[0031] Set the expected false alarm rate, and calculate the adaptive change detection threshold based on the test statistic and the false alarm rate.

[0032] Optionally, in one embodiment of this application, the test statistic is f. c (U,V|H0), the adaptive change detection threshold is:

[0033]

[0034] in, Let P be the detection threshold to be determined, ρ be the obtained adaptive detection threshold, and P be the detection threshold to be determined. FA The set expected false alarm rate.

[0035] Optionally, in one embodiment of this application, the test statistic of the remote sensing image before and after the change to be detected is calculated using a trained Copula function, including:

[0036] Calculate the set of superpixel feature values ​​for all superpixels in each image;

[0037] Perform a probability integral transformation on the calculated set of superpixel feature values ​​to obtain the feature values ​​of the probability distribution function corresponding to the probability integral transformation.

[0038] The eigenvalues ​​of the probability distribution function are input into the trained Copula function to obtain the test statistic;

[0039] The detection threshold value is used to detect the test statistic, and a change detection result is obtained, including:

[0040] The detection threshold value is used to detect the test statistic, and a change detection result is obtained, including:

[0041] Optionally, in an embodiment of the present application, the detection threshold value is used to detect the test statistic, and is expressed as:

[0042]

[0043] Wherein, each element in Z corresponds to a superpixel pair in remote sensing images X and Y, and p represents the adaptive detection threshold value.

[0044] To achieve the above purpose, the second aspect of the embodiment of the present application provides a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, when the processor executes the computer program, the detection threshold value adaptive change detection method is realized.

[0045] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0046] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the accompanying drawings, in which:

[0047] Figure 1 The flowchart of the detection threshold value adaptive change detection method provided by the first embodiment of the present application is shown in the figure.

[0048] Figure 2 The flowchart of the detection threshold value adaptive change detection method provided by the second embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0049] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the drawings, wherein the same or similar reference signs represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0050] The detection threshold value adaptive change detection method and device of the embodiments of the present application are described below with reference to the drawings.

[0051] Figure 1A flowchart of a change detection method with adaptive detection threshold provided by Embodiment One of the present application is shown.

[0052] As shown in Figure 1 , the change detection method with adaptive detection threshold comprises the following steps:

[0053] Step 101, performing collaborative superpixel segmentation on the remote sensing images before and after the change to obtain the superpixel segmentation results of the remote sensing images before and after the change;

[0054] Step 102, calculating the affinity matrix of the remote sensing images before and after the change;

[0055] Step 103, selecting superpixel-level training samples through the superpixel segmentation results and the affinity matrix of the remote sensing images before and after the change;

[0056] Step 104, training the Copula function through the superpixel-level training samples, and generating an adaptive change detection threshold using the trained Copula function;

[0057] Step 105, calculating the test statistics of the remote sensing images before and after the change through the trained Copula function, and performing detection judgment on the test statistics using the adaptive change detection threshold to obtain the change detection result.

[0058] The change detection method with adaptive detection threshold of the present application introduces an adaptive threshold setting mechanism based on Copula theory, dynamically adjusts the threshold by analyzing the marginal distribution and joint distribution of multi-temporal remote sensing images to adapt to the change characteristics in different scenarios. Specifically, first, the multi-temporal remote sensing images are preprocessed, features are extracted, and a marginal distribution model is established, then the dependence relationship between the images is modeled using the Copula function, and a suitable change detection threshold is generated by combining statistical learning methods. This adaptive mechanism can effectively reduce false positives and false negatives, improve the accuracy of change detection, and maintain robustness in various complex environments. The present application not only improves the performance of change detection, but also provides a more reliable theoretical basis and practical guidance for the analysis of remote sensing images.

[0059] Please refer to Figure 2 , Figure 2 A flowchart of a change detection method with adaptive detection threshold provided by Embodiment Two of the present application is shown. The method is executed by an electronic device.

[0060] Let and be a pair of registered remote sensing images of the same area obtained at different time points T1 and T2, The image channel number of two images X and Y respectively, the target of change detection is to find out the changed area in X and Y, and generate a pixel-level binary detection map.

[0061] As shown in the embodiment, the flow of change detection is as follows: Figure 2

[0062] Step 1: Dimensionality reduction processing is performed on the input remote sensing image

[0063] Firstly, the input image X and image Y are preprocessed, and the steps of preprocessing are as follows: firstly, it is judged whether the channel number of the input image is greater than 1. If the channel number of the image is equal to 1, it can be jumped to step 2. If the channel number of the image is greater than 1, the channel dimensionality reduction processing needs to be performed on the image, and the specific dimensionality reduction method depends on the type of the image. Generally, it can be applied by principal component analysis (PCA), linear discriminant analysis (LDA), independent component analysis (ICA) or non-negative matrix factorization (NMF), or it can be selected according to specific application. After the dimensionality reduction processing is completed, the generated low channel dimension image will have a lower data amount, which is convenient for subsequent image processing, feature extraction and analysis tasks, improves the calculation efficiency and reduces the processing complexity.

[0064] Step 2: Cooperative superpixel segmentation is performed on the image

[0065] In the proposed method, a cooperative segmentation strategy based on SLIC (Simple Linear Iterative Clustering) simple linear iterative clustering algorithm is used. Firstly, a reference image needs to be selected for superpixel segmentation. Considering the integrity of superpixel segmentation and the accuracy of feature boundary description, an optical remote sensing image with higher resolution can be selected as the reference image. An initial superpixel center is initialized for each superpixel, and clustering is performed in the color space of the image. Each superpixel center represents an initial clustering center, which is uniformly distributed in the color space and spatial coordinates of the image. Then, the algorithm updates each pixel to its nearest superpixel center through iteration until the boundary of the superpixel tends to be stable. In each iteration, SLIC calculates the distance between each pixel and its adjacent superpixel center, which is measured by considering the weighted sum of color difference and spatial distance. SLIC adjusts the boundary of the superpixel in this way to better adapt to the texture and color changes of the image. After several iterations, the superpixels will gradually tend to be stable, and each superpixel will have higher uniformity and consistency. It is assumed that and are two superpixel sets after SLIC algorithm segmentation of the double-time-phase remote sensing image, the number of superpixels in the two superpixel sets is N1 and N2 respectively, is the i-th pixel segmented from the H-phase image, and ​is the jth pixel segmented from the Y temporal image. Let the corresponding and superpixels intersect to form a new independent segmentation region, denoted as By eliminating the empty set in Λ S and merging small regions (i.e. smaller than the minimum region in N s to the nearest larger region, a complete segmentation map can be further obtained. Where the ith superpixels in X and Y can be represented as:

[0066]

[0067] Step 3: Select superpixel-level training samples

[0068] After the superpixel segmentation is completed, some unchanged superpixel pairs are selected as training samples. The selection method is as follows: First, calculate an affinity matrix for each remote sensing image. Then, calculate the difference between the affinity matrices of each remote sensing image to obtain a change probability map. The change probability map reflects the probability of change of each pixel point on the remote sensing image. Next, intersect the superpixel map segmented in step 2 with the change probability map to obtain the change probability set of the pixel points within each superpixel pair. The average value of the elements in the change probability set of the pixel points is obtained, which is the change probability of each superpixel pair. Finally, sort the change probabilities of all superpixels, and select the top 20% with smaller change probabilities as training superpixels.

[0069] Step 4: Train the Copula function

[0070] Before training the Copula function, the average pixel intensity within each selected training superpixel is calculated to obtain the superpixel feature value of the superpixel. For each selected training superpixel on the time corresponding image, the feature values of the superpixels will form a training superpixel feature value set, denoted as M and N) respectively. Next, perform probability integral transformation on the training superpixel feature value set on each time corresponding image to transform the training superpixel feature value into a probability distribution function value feature. Formally expressed as:

[0071] U = F M (M)

[0072] V = F N (N)

[0073] Where F M and F Nare the probability distribution functions of M and N respectively. Next, the probability distribution function value features of multiple time corresponding images are input into the Copula function for training fitting, and the correlation between the probability distribution function value features is extracted. It is expressed as:

[0074]

[0075] wherein θ represents the internal parameter of the current Copula function, for the convenience of expression, in the following, the is no longer explicitly written, represents the optimal parameter found.

[0076] The best parameter of the Copula function finally trained is

[0077] Step 5: Adaptive detection threshold generation

[0078] When the training of the Copula function is completed, the adaptive threshold is generated by using the trained Copula function. Next, the change detection can be converted into a binary hypothesis testing problem by using the trained Copula function, and the given binary hypothesis is change and no change, let H0 be the no change hypothesis, H1 be the change hypothesis, the joint distribution function of the probability distribution function value features U and V under the no change hypothesis and the change hypothesis is Copula(U, V|H0), Copula(U, V|H1) respectively. The density function of the probability distribution function value features U and V under the no change hypothesis and the change hypothesis is f c (U, V|H0), f c (U, V|H1) respectively. Under the change hypothesis, the different time corresponding remote sensing image regions should obey independent distribution, which means that the Copula function between the different time remote sensing image regions under the change hypothesis should be 1, that is, f c (U, V|H1) = 1. Under the no change hypothesis, the distribution function obeyed between the different time corresponding remote sensing images is f c (U, V|H0), according to the Neyman-Pearson criterion, under the condition of a certain false positive rate, the maximum detection rate can be obtained by using the likelihood ratio test method. Therefore, the likelihood ratio test is used in this embodiment, and the test statistic obtained is f c (U, V|H0). At this time, a desired false positive rate P FA can be obtained by the following formula:

[0079]

[0080] Step 6: Change detection result reasoning

[0081] After obtaining the adaptive threshold p, first, the superpixel feature value set of all selected superpixels on the image corresponding to each time is calculated, and then the obtained superpixel feature value set is subjected to a probability integral transform to obtain the probability distribution function value features corresponding to X and Y after the probability integral transform and and In the input step 4, the obtained test statistic is denoted as Z. The adaptive threshold p is used for detection and judgment on Z, that is,

[0082]

[0083] Each element in Z corresponds to a superpixel pair in the original image X and Y. The detection result is traced back to the pixel level, and the binary change detection result is output.

[0084] In order to realize the above-mentioned embodiments, the application further provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to realize the method described in the above-mentioned embodiments.

[0085] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in combination with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0086] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0087] ​Any processes or methods described in the flowcharts or otherwise described herein can be understood as representing modules, segments, or portions of code that include one or more executable instructions for implementing specific logical functions or steps, and the various embodiments of the application can include additional or fewer steps performing the same or equivalent functions as those shown or discussed, in different orders, including substantially simultaneous execution of the functions described with respect to particular steps, and the like.

[0088] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing the logic function, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or a combination thereof. For the purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can specifically include the following, which are non-exhaustive listings: electrical connections (electrical apparatus) having one or more wires, a portable computer diskette (magnetic apparatus), a random-access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, since the program can be electronically captured, for example, by optically scanning the paper or other medium, then electronically converted into a form that is suitable for use in a computer storage medium.

[0089] It should be understood that portions of the application can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. As such, if implemented in hardware, and in another embodiment, any of the following technologies, known in the art, or a combination thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0090] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment method can be completed by programs instructing related hardware, and the programs can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0091] In addition, each functional unit in each embodiment of the present application can be integrated into one processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0092] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for detecting changes in a detection threshold adaptively, characterized in that, include: Cooperative superpixel segmentation is performed on the remote sensing images before and after the change to obtain the superpixel segmentation results of the remote sensing images before and after the change. Calculate the affinity matrix of the remote sensing images before and after the change; Superpixel-level training samples are selected based on the superpixel segmentation results and affinity matrix of the remote sensing images before and after the changes. The Copula function is trained using superpixel-level training samples, and the trained Copula function is used to generate an adaptive change detection threshold. The test statistics of the remote sensing images before and after the change are calculated by the trained Copula function, and the change detection result is obtained by using an adaptive change detection threshold to make a detection decision on the test statistics. The superpixel-level training samples include training superpixels from remote sensing images before and after the changes. Training the Copula function using the superpixel-level training samples includes: The superpixel feature value of each superpixel is obtained by averaging the pixel intensity within each training superpixel. The superpixel feature values ​​of all training superpixels corresponding to each remote sensing image are combined to form the training superpixel feature value set of that remote sensing image. A probability integral transformation is performed on the training superpixel feature value set of each remote sensing image to transform the training superpixel feature values ​​into probability distribution function value features; The probability distribution function values ​​of the remote sensing images before and after the change are input into the Copula function for training and fitting. The correlation between the probability distribution function values ​​is extracted to obtain the optimal parameters of the Copula function. The remote sensing images before and after the change are Let X and Y be the number of image channels, where X and Y are paired remote sensing images of the same region obtained at different time points T1 and T2 after registration. The training superpixel feature sets corresponding to X and Y are M and N, respectively, and their probability distribution function values ​​are: U=F M (M) V=F N (N) Among them, F M Let F be the probability distribution function of M. N Let N be the probability distribution function; The optimal parameters for the Copula function are: in, The optimal parameters found are represented by θ, which represents the internal parameters of the current Copula function. The Copula density of the Copula function is represented by Copula'(U,V), and the trained Copula is denoted by θ. In If not explicitly written out, it is represented as Copula(U,V); The step of generating an adaptive change detection threshold using a trained Copula function includes: The change detection problem is transformed into a binary hypothesis testing problem using a trained Copula function. The binary hypotheses include a change hypothesis and a no-change hypothesis. Let H1 be the change hypothesis and H0 be the no-change hypothesis. The Copula function is the joint distribution function of the value features U and V under the no-change and change hypotheses, respectively: Copula(U,V|H0) and Copula(U,V|H1). The Copula function is the probability density function of the value features U and V under the no-change and change hypotheses, respectively: fi. c (U,V|H0), f c (U,V|H1), under the assumption of variation, the remote sensing image regions corresponding to different times follow independent distributions, and the probability density function of the Copula function of the remote sensing image regions corresponding to different times is 1, expressed as: f c Given (U,V|H1)=1, and assuming no changes, the Copula density between remote sensing images at different times is f. c (U,V|H0); According to the Neyman-Pearson criterion, given a fixed false alarm rate, the likelihood ratio test is used to obtain the test statistic. Set the expected false alarm rate, and calculate the adaptive change detection threshold based on the test statistic and the false alarm rate; The test statistic is f. c (U,V|H0), the adaptive change detection threshold is: in, Let P be the detection threshold to be determined, ρ be the obtained adaptive detection threshold, and P be the detection threshold to be determined. FA The set expected false alarm rate.

2. The method as described in claim 1, characterized in that, Collaborative superpixel segmentation of remote sensing images is achieved using the SLIC simple linear iterative clustering algorithm; Before performing collaborative superpixel segmentation on the remote sensing images before and after the changes, the method further includes: preprocessing the remote sensing images, wherein the preprocessing process includes: Determine if the number of channels in the remote sensing image is greater than 1. If it is, perform channel dimensionality reduction processing on the remote sensing image.

3. The method as described in claim 1, characterized in that, The superpixel segmentation result is a segmented superpixel image. The selection of superpixel-level training samples based on the superpixel segmentation results and affinity matrix of the remote sensing images before and after the change includes: The differences between the affinity matrices of each remote sensing image are calculated to obtain a probability map of change. The segmented superpixel map is intersected with the change probability map to obtain the set of change probabilities of pixels within each pair of superpixels; The average value of the elements in the set of pixel change probabilities is used to obtain the change probability of each pair of superpixels. According to the priority of the change probability from small to large, a preset proportion of superpixels are selected from all superpixels as training superpixels, and the selected training superpixels are used as the superpixel-level training samples.

4. The method as described in claim 1, characterized in that, The calculation of the test statistics of the remote sensing images before and after the changes to be detected using the trained Copula function includes: Calculate the set of superpixel feature values ​​for all superpixels in each image; Perform a probability integral transformation on the calculated set of superpixel feature values ​​to obtain the feature values ​​of the probability distribution function corresponding to the probability integral transformation. The eigenvalues ​​of the probability distribution function are input into the trained Copula function to obtain the test statistic; The step of using an adaptive change detection threshold to detect and decide on the test statistic to obtain the change detection result includes: An adaptive threshold is used to make a detection decision on the test statistic. When making the detection decision, the detection result is traced back to the pixel level to obtain the binary change detection result.

5. The method as described in claim 4, characterized in that, Using an adaptive change detection threshold to make a decision on the test statistic, it is expressed as: In this test statistic Z, each element corresponds to a superpixel pair in the remote sensing images X and Y, and ρ represents the adaptive detection threshold.

6. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the method as described in any one of claims 1-5.

Images