Demixing-based end-to-end hyperspectral image change detection method and system

Abstract

The invention relates to the field of remote sensing image processing, and discloses an end-to-end hyperspectral image change detection method and system based on unmixing, and the method comprises the steps: obtaining a to-be-detected dual-time-phase hyperspectral image; subtracting the first time-phase hyperspectral image from the second time-phase hyperspectral image to obtain a to-be-detected image difference image; inputting a to-be-detected image difference map into the trained hyperspectral image change detection network to obtain a dual-time-phase hyperspectral image change detection result; wherein the trained hyperspectral image change detection network comprises a de-mixing sub-network and a change detection sub-network; the demixing sub-network is used for processing a to-be-detected image difference image to obtain an abundance difference image; and the change detection sub-network processes the abundance difference image and the to-be-detected image difference image to obtain a dual-time-phase hyperspectral image change detection result, so that the change detection precision is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing image processing and computer vision technology, and in particular to an end-to-end hyperspectral image change detection method and system based on unmixing. Background Art

[0002] Hyperspectral remote sensing image change detection technology identifies the temporal evolution of surface cover by analyzing multi-temporal imaging data of the same area. Different from conventional multispectral remote sensing image change detection, this technology makes full use of the rich spectral information of hyperspectral remote sensing images, can capture subtle differences in material composition, and has unique value in identifying micro-changes such as vegetation succession and pollutant migration. In the existing technical system, the change detection method based on spectral unmixing effectively improves the ability to identify complex surface changes by analyzing the proportion of ground objects in mixed pixels. The technical principle is that each pixel of a hyperspectral image can be represented as a linear or nonlinear combination of several endmember spectra. By extracting the difference in endmember abundance in dual-temporal images, the temporal changes in the composition of surface materials can be revealed, overcoming the problem of misjudgment of mixed pixels by traditional methods.

[0003] There are two key problems in the existing technology: First, the change detection framework based on unmixing generally adopts a staged processing mode, that is, first extracting the abundance map of the remote sensing image, and then performing change detection on the abundance map. This split processing method makes the unmixing process lack the constraint of the change detection results, resulting in the small errors in the unmixing process being amplified in the change detection process, causing error accumulation and affecting the change detection results. Second, most change detection methods based on unmixing only use abundance features for discrimination, ignoring the complementary information contained in the original hyperspectral image, which limits the improvement of change detection accuracy. Summary of the invention

[0004] In order to solve the shortcomings of the prior art, the present invention provides an end-to-end hyperspectral image change detection method and system based on unmixing; the unmixing network and the change detection network are combined to form an end-to-end network so that they can be trained collaboratively. The unmixing network is responsible for extracting the abundance map, and the change detection network is responsible for fusing the information of the abundance map and the original image for change detection.

[0005] On the one hand, an end-to-end hyperspectral image change detection method based on unmixing is provided, including: Acquire a dual-phase hyperspectral image to be detected; the dual-phase hyperspectral image includes: a first-phase hyperspectral image and a second-phase hyperspectral image; Subtracting the first phase hyperspectral image from the second phase hyperspectral image to obtain an image difference image to be detected; The image difference map to be detected is input into the trained hyperspectral image change detection network to obtain the dual-phase hyperspectral image change detection result; wherein the trained hyperspectral image change detection network includes: a demixing subnetwork and a change detection subnetwork; the demixing subnetwork processes the image difference map to be detected to obtain an abundance difference map; the change detection subnetwork processes the abundance difference map and the image difference map to be detected to obtain the dual-phase hyperspectral image change detection result.

[0006] On the other hand, an end-to-end hyperspectral image change detection system based on unmixing is provided, including: An acquisition module is configured to: acquire a dual-phase hyperspectral image to be detected; the dual-phase hyperspectral image includes: a first-phase hyperspectral image and a second-phase hyperspectral image; A subtraction module is configured to: subtract the first phase hyperspectral image from the second phase hyperspectral image to obtain an image difference image to be detected; The image difference map to be detected is input into the trained hyperspectral image change detection network to obtain the dual-phase hyperspectral image change detection result; wherein the trained hyperspectral image change detection network includes: a demixing subnetwork and a change detection subnetwork; the demixing subnetwork processes the image difference map to be detected to obtain an abundance difference map; the change detection subnetwork processes the abundance difference map and the image difference map to be detected to obtain the dual-phase hyperspectral image change detection result.

[0007] The above technical solution has the following advantages or beneficial effects: Through the integrated design of the unmixing network and the change detection network, dual-task collaborative optimization is achieved. The unmixing process directly serves the change detection task, solving the error accumulation problem in the traditional step-by-step method; secondly, an information fusion module is designed to dynamically fuse the original hyperspectral image information with the abundance information of the abundance map using the cross-attention mechanism, alleviating the limitations of single-modality data. This scheme uses end-to-end training to automatically adapt the abundance estimation to the change detection requirements, significantly improving the change detection accuracy and providing a more advanced solution for change detection in complex mixed pixel scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0009] Figure 1 This is a flow chart of the method of embodiment 1. DETAILED DESCRIPTION

[0010] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0011] Embodiment 1 This embodiment provides an end-to-end hyperspectral image change detection method based on unmixing; like Figure 1 As shown, the end-to-end hyperspectral image change detection method based on unmixing includes: S101: Acquire a dual-phase hyperspectral image to be detected; the dual-phase hyperspectral image includes: a first-phase hyperspectral image and a second-phase hyperspectral image; S102: subtracting the first phase hyperspectral image from the second phase hyperspectral image to obtain an image difference image to be detected; S103: Inputting the image difference map to be detected into the trained hyperspectral image change detection network to obtain a dual-phase hyperspectral image change detection result; wherein the trained hyperspectral image change detection network includes: a demixing subnetwork and a change detection subnetwork; the demixing subnetwork processes the image difference map to be detected to obtain an abundance difference map; the change detection subnetwork processes the abundance difference map and the image difference map to be detected to obtain a dual-phase hyperspectral image change detection result.

[0012] Furthermore, the step S101: acquiring the dual-phase hyperspectral image to be detected is performed by using an imaging spectrometer.

[0013] Furthermore, the dual-phase hyperspectral image includes: a first-phase hyperspectral image and a second-phase hyperspectral image, wherein the acquisition time of the first-phase hyperspectral image is earlier than the acquisition time of the second-phase hyperspectral image.

[0014] Further, S102: subtracting the first phase hyperspectral image from the second phase hyperspectral image to obtain an image difference image to be detected, including:

[0015] in, Indicates taking the absolute value, represents the image difference map to be detected, represents the first phase hyperspectral image, Represents the hyperspectral image of the second phase.

[0016] Furthermore, in the actual application stage, the demixing subnetwork processes the image difference map to be detected to obtain an abundance difference map, specifically including: (11) The image difference map to be detected Input to the unmixing subnetwork for unmixing processing to obtain the abundance difference map ; The specific process is as follows: First, the convolution kernel size is 3 3's two-dimensional convolution extracts image features and reduces the image dimension, and then passes through the ReLU activation function to further constrain its data range to obtain intermediate features . The image difference map The number of channels is C. After convolution processing, the number of image channels is reduced to C / / 2, where / / represents integer division; ; (12) Obtaining intermediate features After that, Perform convolution again with a kernel size of 3 3, further reducing its dimension to the same as the number of endmembers P; after convolution, the number of image channels is reduced from C / / 2 to P; then the output is constrained by the Softmax activation function to ensure that the value of each pixel in the abundance difference map is between 0 and 1, and the abundance difference map is obtained. ; .

[0017] Furthermore, in the actual application stage, the change detection subnetwork processes the abundance difference map and the image difference map to be detected to obtain a dual-phase hyperspectral image change detection result, including: (11): The change detection subnetwork extracts features from the image difference map and the abundance difference map respectively to obtain the image difference Figure 1 Sub-treatment characteristics and abundance differences Figure 1 Secondary processing characteristics; (12): Image difference Figure 1 Sub-treatment characteristics and abundance differences Figure 1 The secondary processing features are extracted respectively to obtain the secondary processing features of the image difference map and the secondary processing features of the abundance difference map; (13): Use the cross attention module to process the secondary processing features of the image difference map and the secondary processing features of the abundance difference map to obtain the fusion features; (14) Classify the fused features to obtain the final classification result.

[0018] Furthermore, the change detection subnetwork extracts features from the image difference map and the abundance difference map respectively to obtain the image difference Figure 1 Sub-treatment characteristics and abundance differences Figure 1 Secondary processing characteristics include: The input of the change detection subnetwork is the image difference map Abundance Difference Map , which contain C and P image channels respectively. and The dimensions of and Perform dimension alignment. The specific process is as follows: Image difference map After convolution, the kernel size is 1 After the 2D convolution of 1, the image size remains unchanged, and the number of channels is reduced from C to 64, and the image difference is obtained. Figure 1 Secondary processing characteristics ; Abundance difference map After convolution, the kernel size is 1 After the 2D convolution of 1, the image size remains unchanged, and the number of channels is increased from P to 64, and the abundance difference is obtained. Figure 1 Secondary processing characteristics : ; ; in, Indicates the use of 1 1 The two-dimensional convolution operation of the convolution kernel is used to adjust the number of channels of the input feature map. In this way, the present invention ensures and The consistency in the number of channels provides a unified feature representation for subsequent feature extraction and fusion operations.

[0019] Furthermore, the (12): image difference Figure 1 Sub-treatment characteristics and abundance differences Figure 1 The secondary processing features are extracted respectively to obtain the secondary processing features of the image difference map and the secondary processing features of the abundance difference map, which specifically include: After dimensional alignment, a 2D convolution with a kernel size of 3 × 3 is used. and Perform feature extraction to further obtain its spatial information. The specific process is as follows: After a two-dimensional convolution, the image size and number of channels remain unchanged, and the secondary processing features of the image difference map are obtained. ; After a two-dimensional convolution, the image size and number of channels remain unchanged, and the secondary processing features of the abundance difference map are obtained. :

[0020]

[0021] in, Indicates the use of 3 3 convolution kernel two-dimensional convolution operation, in this way, the present invention not only ensures and The consistency in the number of channels further enhances their spatial feature representation and provides richer information for subsequent feature fusion and change detection.

[0022] Furthermore, the (13): using a cross attention module to process the image difference map secondary processing features and the abundance difference map secondary processing features to obtain fusion features, specifically including: Image difference map secondary processing features It has rich spatial and spectral information, and the abundance difference map has secondary processing characteristics. The two have similar structures and complementary information. Therefore, the Cross Attention module is used to The complementary information in In the above example, we obtain the fusion features The specific process is as follows: First, Input into the fully connected layer Linear1 to calculate the query vector (Query); Input to the fully connected layer Linear2 and fully connected layer Linear3, and calculate the key vector (Key) and value vector (Value) respectively for subsequent attention distribution calculation. The above process can be formulated as:

[0023] Then, the query vector Q is multiplied by the transpose of the value vector K and passed through the SoftMax activation function to obtain and Attention distribution of complementary information The specific calculation formula is as follows:

[0024] in, is the scaling factor of the feature dimension to maintain numerical stability.

[0025] Attention distribution to obtain complementary information Afterwards, by With value vector Multiply by and get the secondary processing feature of the image difference map Secondary processing features with abundance difference map Complementary information between , the specific process can be formulated as:

[0026] in, Represents matrix multiplication; Finally, the complementary information Secondary processing features with image difference map Add together to get the final fusion feature , the above process can be formulated as:

[0027] In this way, the present invention not only retains the spatial and spectral information of the image difference map, but also incorporates the complementary information between it and the abundance difference map, which helps to improve the accuracy and robustness of change detection.

[0028] Furthermore, the (14) classifies the fused features to obtain the final classification result, including: In obtaining fusion features After that, the classification layer is used to classify each pixel in the image. Since the final result of change detection needs to classify the pixels into two categories (i.e., changed pixels and unchanged pixels), it is necessary to reduce the dimension of the features through the pooling layer and the fully connected layer to adapt to the subsequent classification processing. The above process can be formulated as:

[0029] in, The function is to perform global average pooling on the feature map to aggregate global information; Represents a fully connected layer, which is used to adjust the feature dimension.

[0030] Furthermore, in S103, the image difference map to be detected is input into the trained hyperspectral image change detection network to obtain a dual-phase hyperspectral image change detection result; wherein the training process of the trained hyperspectral image change detection network includes: Constructing a data set, the data set is a bi-phase hyperspectral image with known change detection results; subtracting the bi-phase hyperspectral images and taking the absolute value of the subtraction result to obtain a hyperspectral image difference map; extracting endmembers from the hyperspectral image difference map to obtain endmembers; dividing the data set into a training set and a validation set; The training set is input into the hyperspectral image change detection network to train the network. When the total loss function value of the network no longer decreases, the training is stopped to obtain the trained hyperspectral image change detection network. The trained hyperspectral image change detection network is verified using the validation set. If the verification passes, the final trained hyperspectral image change detection network is obtained. If the verification fails, the data set is replaced and the training is repeated. The total loss function is the sum of the cosine loss function and the cross entropy loss function; Assuming that the training includes a total of 2N rounds of training, during the first N rounds of training, the value of the cross entropy loss function decreases slower than the value of the cosine loss function; during the last N rounds of training, the value of the cosine loss function decreases slower than the value of the cross entropy loss function.

[0031] Furthermore, the subtracting of the two-phase hyperspectral images and taking the absolute value of the subtraction result to obtain a hyperspectral image difference map includes:

[0032] in, Indicates taking the absolute value, represents the hyperspectral image difference map, represents the third phase hyperspectral image, Represents the hyperspectral image of the fourth phase.

[0033] Furthermore, the endmember extraction is performed on the hyperspectral image difference map to obtain the endmembers, and a vertex component analysis algorithm is used to implement the endmember extraction.

[0034] It should be understood that first, the dual-phase hyperspectral images X1 and X2 are subtracted and the absolute value is taken to obtain a hyperspectral image difference map. The specific calculation formula is: Then, the endmembers E of the difference graph are extracted and saved through the Vertex Component Analysis (VCA) algorithm for input into the subsequent network. VCA is an unsupervised endmember extraction algorithm based on convex geometry theory. It identifies the spectral features representing the pure object category from the hyperspectral image by iteratively searching for the maximum projection vector.

[0035] It should be understood that the present invention cuts the dual-phase hyperspectral image into a size of 7 7 image patches are used as the input of the network. Small patches can ensure that they contain rich spatial information while avoiding interference from weakly correlated pixels. Then 9% of the data is randomly selected for training, 1% for validation, and the rest for testing.

[0036] Furthermore, during the training process, the demixing network is used to process the image difference map to obtain the abundance difference map, which specifically includes: (21) Image difference map Input to the unmixing subnetwork for unmixing processing to obtain the abundance difference map ; The specific process is as follows: First, the convolution kernel size is 3 3's two-dimensional convolution extracts image features and reduces the image dimension, and then passes through the ReLU activation function to further constrain its data range to obtain intermediate features . The image difference map The number of channels is C. After convolution processing, the number of image channels is reduced to C / / 2, where / / represents integer division; ; (22) Obtaining intermediate features After that, Perform convolution again with a kernel size of 3 3, further reducing its dimension to the same as the number of endmembers P; after convolution, the number of image channels is reduced from C / / 2 to P; then the output is constrained by the Softmax activation function to ensure that the value of each pixel in the abundance difference map is between 0 and 1, and the abundance difference map is obtained. ; .

[0037] Finally, the abundance difference map Multiply it with the extracted end member E to get the reconstructed image difference map , thereby calculating the image difference map Difference map with reconstructed image The cosine similarity loss function between , the above process can be formulated as:

[0038] Cosine similarity loss function , the specific expression is as follows:

[0039] in represents the abundance difference map matrix, represents the endmember matrix, represents matrix multiplication, Represents the first element values, Represents the first element values, and C represents the vector dimension.

[0040] Furthermore, during the training process, the change detection subnetwork processes the abundance difference map and the image difference map to obtain the change detection results of the dual-phase hyperspectral image, including: The input of the change detection subnetwork is the image difference map Abundance Difference Map , which contain C and P image channels respectively. and The dimensions of and Perform dimension alignment. The specific process is as follows: Image difference map After convolution, the kernel size is 1 After the 2D convolution of 1, the image size remains unchanged, and the number of channels is reduced from C to 64, and the image difference is obtained. Figure 1 Secondary processing characteristics ; Abundance difference map After convolution, the kernel size is 1 After the 2D convolution of 1, the image size remains unchanged, and the number of channels is increased from P to 64, and the abundance difference is obtained. Figure 1 Secondary processing characteristics : ; ; in, Indicates the use of 1 1 The two-dimensional convolution operation of the convolution kernel is used to adjust the number of channels of the input feature map. In this way, the present invention ensures and The consistency in the number of channels provides a unified feature representation for subsequent feature extraction and fusion operations; After dimensional alignment, a convolution kernel size of 3 is used. 3 two-dimensional convolution pair and Feature extraction is performed to further obtain its spatial information. The specific process is as follows: After a two-dimensional convolution, the image size and number of channels remain unchanged, and the secondary processing features of the image difference map are obtained. ; After a two-dimensional convolution, the image size and number of channels remain unchanged, and the secondary processing features of the abundance difference map are obtained. :

[0041]

[0042] in, Indicates the use of 3 3 convolution kernel two-dimensional convolution operation, in this way, the present invention not only ensures and The consistency in the number of channels further enhances their spatial feature representation and provides richer information for subsequent feature fusion and change detection.

[0043] After the above processing, the image difference map secondary processing features It has rich spatial and spectral information, and the abundance difference map has secondary processing characteristics. The two have similar structures and complementary information. Therefore, the Cross Attention module is used to The complementary information in In the above example, we obtain the fusion features The specific process is as follows: First, Input into the fully connected layer Linear1 to calculate the query vector (Query); Input to the fully connected layer Linear2 and fully connected layer Linear3, and calculate the key vector (Key) and value vector (Value) respectively for subsequent attention distribution calculation. The above process can be formulated as:

[0044] Then, the query vector Q is multiplied by the transpose of the value vector K and passed through the SoftMax activation function to obtain and Attention distribution of complementary information The specific calculation formula is as follows:

[0045] in, is the scaling factor of the feature dimension to maintain numerical stability.

[0046] Attention distribution to obtain complementary information Afterwards, by With value vector Multiply by and get the secondary processing feature of the image difference map Secondary processing features with abundance difference map Complementary information between , the specific process can be formulated as:

[0047] in, Represents matrix multiplication; Finally, the complementary information Secondary processing features with image difference map Add together to get the final fusion feature , the above process can be formulated as:

[0048] In this way, the present invention not only retains the spatial and spectral information of the image difference map, but also incorporates the complementary information between it and the abundance difference map, which helps to improve the accuracy and robustness of change detection.

[0049] In obtaining fusion features After that, the classification layer is used to classify each pixel in the image. Since the final result of change detection needs to classify the pixels into two categories (i.e., changed pixels and unchanged pixels), it is necessary to reduce the dimension of the features through the pooling layer and the fully connected layer to adapt to the subsequent classification processing. The above process can be formulated as:

[0050] in, The function is to perform global average pooling on the feature map to aggregate global information; Represents a fully connected layer, which is used to adjust the feature dimension.

[0051] Finally, the Softmax activation function is used to obtain the predicted label of whether each pixel has changed.

[0052]

[0053] The cross-entropy loss function is used to measure the difference between the predicted label and the true label. The expression of the cross-entropy loss function is as follows:

[0054] in represents the number of labeled samples, and Respectively represent The true labels and predicted labels of samples.

[0055] Cosine loss function With the cross entropy loss function Add together to get the final network loss function , the expression is as follows:

[0056] In the early stages of network training, The value of is large, which plays a leading role in network training, and the convergence speed of the demixing network is fast. In the middle and late stages of network training, the demixing network tends to be stable. The value of decreases slightly. Plays a leading role, the convergence speed of the change detection sub-network is accelerated, and the de-mixing sub-network will be in the loss function and The unmixing effect is further improved by fine-tuning under the common constraint of , making the abundance maps of the unchanged areas more similar and the abundance maps of the changed areas more different. This enables the coordinated training of the unmixing sub-network and the change detection sub-network.

[0057] Taking the hyperspectral image difference map as input, the end members are first extracted through the vertex component analysis (VCA) algorithm, and then the abundance difference map is obtained by using the network's demixing subnetwork. Finally, the image difference map and the abundance difference map are sent to the change detection subnetwork of the network to obtain the final change detection result. In the change detection subnetwork, the convolution operation is responsible for aligning the feature dimensions and extracting spatial information, and the cross attention module is responsible for fusing the image difference map and the abundance difference map information. This method can effectively overcome the limitations of using a single feature for change detection, and realizes the collaborative training of demixing and change detection through an end-to-end network, effectively improving the accuracy and flexibility of hyperspectral change detection.

[0058] Embodiment 2 This embodiment provides an end-to-end hyperspectral image change detection system based on unmixing, including: An acquisition module is configured to: acquire a dual-phase hyperspectral image to be detected; the dual-phase hyperspectral image includes: a first-phase hyperspectral image and a second-phase hyperspectral image; A subtraction module is configured to: subtract the first phase hyperspectral image from the second phase hyperspectral image to obtain an image difference image to be detected; The image difference map to be detected is input into the trained hyperspectral image change detection network to obtain the dual-phase hyperspectral image change detection result; wherein the trained hyperspectral image change detection network includes: a demixing subnetwork and a change detection subnetwork; the demixing subnetwork processes the image difference map to be detected to obtain an abundance difference map; the change detection subnetwork processes the abundance difference map and the image difference map to be detected to obtain the dual-phase hyperspectral image change detection result.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An end-to-end hyperspectral image change detection method based on unmixing, characterized in that: include: Acquire a dual-phase hyperspectral image to be detected; The dual-phase hyperspectral image includes: a first-phase hyperspectral image and a second-phase hyperspectral image; Subtracting the first phase hyperspectral image from the second phase hyperspectral image to obtain an image difference image to be detected; The image difference map to be detected is input into the trained hyperspectral image change detection network to obtain the dual-phase hyperspectral image change detection result; wherein the trained hyperspectral image change detection network includes: a demixing subnetwork and a change detection subnetwork; the demixing subnetwork processes the image difference map to be detected to obtain an abundance difference map; the change detection subnetwork processes the abundance difference map and the image difference map to be detected to obtain the dual-phase hyperspectral image change detection result.

2. The end-to-end hyperspectral image change detection method based on unmixing according to claim 1, characterized in that: Subtracting the first phase hyperspectral image from the second phase hyperspectral image to obtain an image difference map to be detected, including: ； in, Indicates taking the absolute value, represents the image difference map to be detected, represents the first phase hyperspectral image, Represents the hyperspectral image of the second phase.

3. The end-to-end hyperspectral image change detection method based on unmixing according to claim 1, characterized in that: The demixing subnetwork processes the image difference map to be detected to obtain an abundance difference map, specifically including: (11) The image difference map to be detected Input to the unmixing subnetwork for unmixing processing to obtain the abundance difference map The specific process is as follows: First, the image features are extracted and the image dimension is reduced through a two-dimensional convolution with a convolution kernel size of 3 × 3, and then the data range is further constrained through the ReLU activation function to obtain the intermediate features , where the image difference map The number of channels is C. After convolution processing, the number of image channels is reduced to C / / 2, where / / represents integer division; ； (12) Obtaining intermediate features After that, A two-dimensional convolution with a kernel size of 3 × 3 is performed again to further reduce its dimension to the same as the number of endmembers P; after convolution, the number of image channels is reduced from C / / 2 to P; then the output is constrained by the Softmax activation function to ensure that the value of each pixel in the abundance difference map is between 0 and 1, and the abundance difference map is obtained. ; 。 4. The end-to-end hyperspectral image change detection method based on unmixing according to claim 1, characterized in that: The change detection subnetwork processes the abundance difference map and the image difference map to be detected to obtain a dual-phase hyperspectral image change detection result, including: (11): The change detection subnetwork extracts features from the image difference map and the abundance difference map respectively, and obtains the image difference map one-time processing features and the abundance difference map one-time processing features; (12): Extract features of the image difference map primary processing features and the abundance difference map primary processing features respectively to obtain image difference map secondary processing features and abundance difference map secondary processing features; (13): Use the cross attention module to process the secondary processing features of the image difference map and the secondary processing features of the abundance difference map to obtain the fusion features; (14) Classify the fused features to obtain the final classification result.

5. The end-to-end hyperspectral image change detection method based on unmixing as claimed in claim 4, characterized in that: The change detection subnetwork extracts features from the image difference map and the abundance difference map respectively, and obtains the first-time processing features of the image difference map and the first-time processing features of the abundance difference map, which specifically include: The input of the change detection subnetwork is the image difference map Abundance Difference Map , which contain C and P image channels respectively. and Perform dimension alignment. The specific process is as follows: Image difference map After convolution, the kernel size is 1 After the 2D convolution of 1, the image size remains unchanged, and the number of channels is reduced from C to 64, and the image difference image is processed once. ; Abundance difference map After convolution, the kernel size is 1 After the 2D convolution of 1, the image size remains unchanged, and the number of channels is increased from P to 64, obtaining the abundance difference map with one-time processing features. : ； ； in, Indicates the use of 1 1 The two-dimensional convolution operation of the convolution kernel is used to adjust the number of channels of the input feature map.

6. The unmixing-based end-to-end hyperspectral image change detection method according to claim 5, characterized in that: Feature extraction is performed on the first processing feature of the image difference map and the first processing feature of the abundance difference map respectively to obtain the second processing feature of the image difference map and the second processing feature of the abundance difference map, which specifically include: After dimensional alignment, a 2D convolution with a kernel size of 3 × 3 is used. and Perform feature extraction, the specific process is as follows: After a two-dimensional convolution, the image size and number of channels remain unchanged, and the secondary processing features of the image difference map are obtained. ; After a two-dimensional convolution, the image size and number of channels remain unchanged, and the secondary processing features of the abundance difference map are obtained. : ； ； in, Indicates the use of 3 2D convolution operation with 3 convolution kernels.

7. The unmixing-based end-to-end hyperspectral image change detection method according to claim 6, characterized in that: The cross attention module is used to process the secondary processing features of the image difference map and the secondary processing features of the abundance difference map to obtain fusion features, including: Using the cross-attention module, The complementary information in In the above example, we obtain the fusion features ; The specific process is as follows: first, Input into the fully connected layer Linear1 to calculate the query vector; Input into the fully connected layer Linear2 and the fully connected layer Linear3 to calculate the key vector and value vector respectively for subsequent attention distribution calculation: ； Then, the query vector Q is multiplied by the transpose of the value vector K and passed through the SoftMax activation function to obtain and Attention distribution of complementary information , the specific calculation formula is as follows: ； in, is the scaling factor of the feature dimension; Attention distribution to obtain complementary information Afterwards, by With value vector Multiply by and get the secondary processing feature of the image difference map Secondary processing features with abundance difference map Complementary information between : ； in, Represents matrix multiplication; Finally, the complementary information Secondary processing features with image difference map Add together to get the final fusion feature : 。 8. The unmixing-based end-to-end hyperspectral image change detection method according to claim 7, characterized in that: The fusion features are classified to obtain the final classification results, including: In obtaining fusion features Finally, the features are reduced in dimension through the pooling layer and the fully connected layer: ； in, The function is to perform global average pooling on the feature map to aggregate global information; Represents a fully connected layer, which is used to adjust the feature dimension.

9. The unmixing-based end-to-end hyperspectral image change detection method according to claim 1, characterized in that: The image difference map to be detected is input into the trained hyperspectral image change detection network to obtain the dual-phase hyperspectral image change detection result; wherein the training process of the trained hyperspectral image change detection network includes: Constructing a data set, the data set is a bi-phase hyperspectral image with known change detection results; subtracting the bi-phase hyperspectral images and taking the absolute value of the subtraction result to obtain a hyperspectral image difference map; extracting endmembers from the hyperspectral image difference map to obtain endmembers; dividing the data set into a training set and a validation set; The training set is input into the hyperspectral image change detection network to train the network. When the total loss function value of the network no longer decreases, the training is stopped to obtain the trained hyperspectral image change detection network. The trained hyperspectral image change detection network is verified using the validation set. If the verification passes, the final trained hyperspectral image change detection network is obtained. If the verification fails, the data set is replaced and the training is repeated. The total loss function is the sum of the cosine loss function and the cross entropy loss function; Assuming that the training includes 2N rounds of training in total, during the first N rounds of training, the value of the cross entropy loss function decreases slower than the value of the cosine loss function; during the last N rounds of training, the value of the cosine loss function decreases slower than the value of the cross entropy loss function; Cosine similarity loss function , the specific expression is as follows: ； in, represents the abundance difference map matrix, represents the endmember matrix, represents matrix multiplication, Represents the first element values, Represents the first spectral vector of the original image element values, C represents the vector dimension; The expression of the cross entropy loss function is as follows: ； in, represents the number of labeled samples, and Respectively represent The true labels and predicted labels of samples.

10. An end-to-end hyperspectral image change detection system based on unmixing, characterized in that: include: An acquisition module is configured to: acquire a dual-phase hyperspectral image to be detected; The dual-phase hyperspectral image includes: a first-phase hyperspectral image and a second-phase hyperspectral image; A subtraction module is configured to: subtract the first phase hyperspectral image from the second phase hyperspectral image to obtain an image difference image to be detected; The image difference map to be detected is input into the trained hyperspectral image change detection network to obtain the dual-phase hyperspectral image change detection result; wherein the trained hyperspectral image change detection network includes: a demixing subnetwork and a change detection subnetwork; the demixing subnetwork processes the image difference map to be detected to obtain an abundance difference map; the change detection subnetwork processes the abundance difference map and the image difference map to be detected to obtain the dual-phase hyperspectral image change detection result.

Images