Hyperspectral image domain change detection method, device and software product

Abstract

This invention discloses a hyperspectral image domain adaptive change detection method, device, and software product. First, it employs correlation alignment loss to minimize image distribution differences caused by variations in imaging time, thereby preventing the network from over-capturing non-critical changes. Simultaneously, this invention introduces a channel attention mechanism to enhance the feature representation of dissimilar images, thus helping the network more accurately detect subtle changes. Finally, this invention uses Kullback-Leibler divergence loss to fuse the features extracted from the two branches, further enhancing the network's accuracy in change detection.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing technology, and relates to a method, device and software product for detecting adaptive changes in the hyperspectral image domain, particularly a method for detecting adaptive changes in the hyperspectral image domain that integrates correlation alignment and spectral attention features. Background Technology

[0002] Change detection identifies and analyzes changes in a specific area by comparing different datasets, and is commonly used in various fields such as disaster assessment, urban planning, and building damage assessment (References 1, 2). Hyperspectral images, due to their rich spectral information, can provide greater decision support for detecting constantly changing geographic features in remote sensing images. Therefore, the application of hyperspectral images in change detection is becoming increasingly widespread, contributing to a deeper understanding and analysis of changes in geographic features.

[0003] Traditional change detection methods can be broadly categorized into three main types (References 3 and 4): First, image algebra-based methods, which effectively detect significant changes in remote sensing images by analyzing spectral differences between pixels; second, transformation-based methods, which capture and extract key changes by mapping the original remote sensing image to a specific feature space; and finally, classification-based methods, which accurately identify changed regions by comparing differences in geographic feature labels within the remote sensing image. However, traditional change detection methods often focus excessively on extracting shallow features, which to some extent limits the comprehensive and in-depth understanding and utilization of remote sensing image information.

[0004] Convolutional neural networks (CNNs) have garnered significant attention due to their powerful ability to extract deep features and have been well-appointed in hyperspectral change detection (references 5, 6, 7). Change detection in hyperspectral images often identifies changes by comparing differences between different datasets. However, due to factors such as imaging time and sensor variations, the resulting hyperspectral images often exhibit differences in feature distribution between images, significantly increasing the complexity of change detection. Therefore, effectively reducing the feature distribution differences between hyperspectral images from different time periods and mitigating the model's over-prediction of change information is the first key challenge in this field. Simultaneously, most methods primarily focus on processing feature differences between hyperspectral images from different time periods, neglecting the inherent differences between the original images. Therefore, deeply exploring the inherent differences between the original images to enhance the model's ability to detect subtle changes is the second key challenge. Furthermore, most networks need further improvement in effectively fusing features extracted from different branches, representing the third key challenge in this field.

[0005] References:

[0006] [1]Sicong Liu,Daniele Marinelli,Lorenzo Bruzzone,and FrancescaBovolo,”Areview of change detection in multitemporal hyperspectral images:Currenttechniques,applications,and challenges,”IEEE Geosci.RemoteSens.Mag.,2019,

[0007] vol.7,no.2,pp:140-158.

[0008] [2]Lorenzo Bruzzone,and Francesca Bovolo,”A novel framework forthedesign ofchange-detection systems for very-high-resolution remotesensingimages,”Proc.

[0009] IEEE,2012,vol.101,no.3,pp:609-630.

[0010] [3]Xiaofeng Zhang,Shuli Cheng,Liejun Wang,and HaojinLi,”Asymmetriccross-attention hierarchical network based on CNN andtransformerforbitemporal remote sensing images change detection,”IEEE Trans.

[0011] Geosci.Remote Sens.,2021,vol.60,pp:1-12.

[0012] [4]Zengfu Hou,Wei Li,Lu Li,Ran Tao,and Qian Du,”Hyperspectralchangedetection based on multiple morphological profiles,”IEEETrans.Geosci.RemoteSens.,2022,vol.60,pp:1-12.

[0013] [5] Meiqi Hu, Chen Wu, and Liangpei Zhang, “HyperNet:

[0014] Self-supervisedhyperspectral spatial spectral feature understanding network for hyperspectralchange detection," IEEE Trans.Geosci.Remote Sens., 2022, vol.60,

[0015] pp:1-17.

[0016] [6]Jigang Ding, Xiaorun Li, Shu Xiang, and Shuhan Chen, "Multilevelfeatures fused and change information enhanced neural network for hyperspectral image change detection," IEEE Trans.Geosci.RemoteSens., 2023.

[0017] [7] Xianghai Wang, Keyun Zhao, Xiaoyang Zhao, and Siyao Li, "TriTF: Atriplettransformer framework based on parents and brother attention for hyperspectral image change detection." IEEE Trans.Geosci.RemoteSens., 2023, vol.61, pp: 1-13. Summary of the Invention

[0018] To address the shortcomings of existing technologies, this invention provides a high-precision and high-efficiency method, device, and software product for detecting hyperspectral image domain adaptive changes by fusing correlation alignment and spectral attention features.

[0019] In a first aspect, the present invention provides a method for detecting hyperspectral image domain adaptation changes, comprising the following steps:

[0020] Step 1: Obtain hyperspectral remote sensing images of the same region at the first time point. Hyperspectral remote sensing images acquired at different times, including the second time point. Where C, H, and W refer to the spectral dimension, length, and width of the hyperspectral remote sensing image, respectively. Based on Siamese networks, the differential depth features F between T1 and T2 are extracted. DF ;

[0021] Step 2: Extract deep features of T1 and T2 based on Siamese network and Furthermore, a correlation alignment loss is constructed based on deep features.

[0022] Step 3: Using a convolutional neural network based on channel attention mechanism, extract the enhanced difference feature F between T1 and T2. DI ;

[0023] Step 4: Transfer the depth features F DF and depth features F DI By directly adding them together, the fused feature FF is obtained;

[0024] Step 5: Construct F based on Kullback-Leibler divergence. DF and F DI of The loss function promotes feature fusion, and the fused features FF are input into the Softmax layer to obtain the change detection results.

[0025] Preferably, in step 1, deep features of T1 and T2 are extracted based on Siamese networks. and Construct differential deep features F using deep features DF ;

[0026] The twin network first processes T1 or T2 respectively through a two-dimensional convolutional neural network and then outputs the results. in, It is the feature output of phase T1 or T2 after processing by a two-dimensional convolutional neural network, W 1 / 2 For convolution weights, B 1 / 2 The bias weights are σ(·), and the activation function is σ(·). Then... and Subtract them and take the absolute value to obtain the difference features. Where |·| represents the absolute value operation; then the difference features are input into a two-dimensional convolutional neural network to extract deep difference features, and finally the deep difference features T are... DF The differential depth features F are obtained by feeding them into a fully connected layer. DF .

[0027] Preferably, in step 2, the correlation alignment loss is based on the deep features of T1 and T2 extracted from the Siamese network. and During training, one of the loss functions used is the domain-adaptive relevance alignment loss function.

[0028]

[0029]

[0030] in, Let n1 represent the square of the Frobenius criterion matrix, 1 represent a column vector consisting entirely of 1s, and n1 and n2 represent the number of samples in the two time phases T1 and T2, respectively.

[0031] Preferably, in step 3, the enhanced difference feature F is extracted based on a convolutional neural network. DI The twin network includes a channel attention mechanism, a two-dimensional convolutional neural network, and a fully connected layer.

[0032] The channel attention mechanism directly subtracts the original images of two time phases, T1 and T2, and performs an absolute value operation to obtain the difference image T. DI And enhance the difference image T based on the channel attention mechanism. DI Dimensional features, to obtain enhanced features Then, a two-dimensional convolutional neural network is used to extract... deep features Finally, the deep features of the enhanced difference image will be... By feeding it into a fully connected layer, we obtain an enhanced difference feature F based on a channel attention mechanism. DI ;

[0033] in, Depend on Reconstructed; Depend on and The result is obtained by transposing and multiplying the matrix. It is a scaling parameter;

[0034] in, It is processed by a 2D convolutional neural network. The feature output is W, where W is the convolution weight, B is the bias weight, and σ(·) is the activation function.

[0035] Preferably, the Siamese network and the convolutional neural network constitute a hyperspectral image domain adaptive change detection network; the hyperspectral image domain adaptive change detection network is a trained network, and the loss function used during training is... for:

[0036]

[0037] in, The cross-entropy loss function for binary classification is... Let the relevance alignment loss function be for domain adaptation. Y represents the Kullback-Leibler divergence loss function for feature fusion; x represents the specific value of each feature; n represents the number of samples; and Y represents the number of samples. i and Represent the true label and predicted label of sample i, respectively; λ1 and λ2 are... and Hyperparameters.

[0038] Secondly, the present invention provides a hyperspectral image domain adaptive change detection device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the hyperspectral image domain adaptive change detection method.

[0039] Thirdly, the present invention provides a software product, including a computer program, which, when executed by a processor, implements the hyperspectral image domain adaptive change detection method.

[0040] Fourthly, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the hyperspectral image domain adaptive change detection method.

[0041] The advantages and beneficial effects of this invention include:

[0042] (1) In order to reduce the distribution differences of hyperspectral images at different time phases, this paper proposes to perform domain adaptation by correlation alignment loss. The aim is to minimize the distribution differences by aligning the correlation features extracted from the hyperspectral images, so as to reduce the model's ability to over-predict changes.

[0043] (2) In order to enhance the network’s ability to mine the features of the original image, this invention enhances the features of the channel dimension of the difference image through the channel attention mechanism, aiming to explore the correlation between the channels of the difference image and further mine semantic features to improve the model’s ability to recognize subtle changes.

[0044] (3) In order to better integrate the different features extracted by the network, this invention reduces the differences in features of different branches by using Kullback-Leibler divergence loss, aiming to promote the integration of features extracted by the two branches and improve the accuracy and reliability of the model in recognizing changing information. Attached Figure Description

[0045] The technical solutions of the present invention will be further illustrated below using embodiments and specific implementation methods. In addition, some accompanying drawings are used in the description of the technical solutions. Those skilled in the art can obtain other drawings and the intent of the present invention from these drawings without any creative effort.

[0046] Figure 1 This is a schematic diagram illustrating the method principle of an embodiment of the present invention. Detailed Implementation

[0047] The following is combined Figure 1 The specific embodiments of the present invention will be further described in detail below.

[0048] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0049] Change detection is a crucial task in geospatial analysis, aiming to identify significant changes in geographic features across images from different datasets. However, existing methods often overlook distributional differences between images caused by temporal variations in imaging. Meanwhile, the features of hyperspectral images still hold great potential for description and detection of change. To address this, this invention proposes a domain-adaptive change detection method for hyperspectral images, combining correlation alignment and spectral attention features. By integrating domain adaptation and attention mechanisms, this method overcomes the shortcomings of existing approaches, thereby achieving more accurate hyperspectral image change detection. Through this method, this invention aims to improve the accuracy, reliability, and efficiency of hyperspectral image change detection, providing strong support for its research and application.

[0050] Please see Figure 1 This embodiment provides a method for detecting hyperspectral image domain adaptation changes, including the following steps:

[0051] Step 1: Obtain hyperspectral remote sensing images of the same region at the first time point. Hyperspectral remote sensing images acquired at different times, including the second time point. Where C, H, and W refer to the spectral dimension, length, and width of the hyperspectral remote sensing image, respectively, and the depth features T1 and T2 are extracted based on Siamese networks. and Finally, differential deep features F are constructed using deep features. DF ;

[0052] In one implementation, deep features of T1 and T2 are extracted based on a Siamese network. and Finally, differential deep features F are constructed using deep features. DF ;

[0053] The twin network first processes T1 or T2 respectively through a two-dimensional convolutional neural network and then outputs the results. in, It is the feature output of phase T1 or T2 after processing by a two-dimensional convolutional neural network, W 1 / 2 For convolution weights, B 1 / 2 The bias weights are σ(·), and the activation function is σ(·), which is ReLU used here. Then... and Subtract them and take the absolute value to obtain the difference features. Where |·| represents the absolute value operation; then the difference features are input into a two-dimensional convolutional neural network to extract deep difference features, and finally the deep difference features T are... DF The differential depth features F are obtained by feeding them into a fully connected layer. DF .

[0054] Step 2: In one implementation, the correlation alignment (CORAL) loss is used to extract deep features of T1 and T2 based on the Siamese network. and During training, one of the loss functions used is the domain-adaptive relevance alignment loss function.

[0055]

[0056] in, Let n1 represent the square of the Frobenius criterion matrix, 1 represent a column vector consisting entirely of 1s, and n1 and n2 represent the number of samples in the two time phases T1 and T2, respectively.

[0057] Step 3: Extract the deep features F of T1 and T2 based on the convolutional neural network. DI ;

[0058] In one implementation, a Siamese network is used to extract deep features F. DI The twin network includes a channel attention mechanism, a two-dimensional convolutional neural network, and a fully connected layer.

[0059] The channel attention mechanism will connect the two time phases T 1 / 2 The images are directly subtracted, and the absolute value is calculated to obtain the difference image T. DI :

[0060] T DI =|T1-T2|;

[0061] To reveal the interdependencies between the channels of the difference image more deeply and to effectively extract the semantic features contained therein, a channel attention mechanism is used to further enhance the feature representation of the difference image:

[0062]

[0063] in, Depend on Reconstructed; Depend on and Obtained by matrix multiplication; It is a scaling parameter;

[0064] Then, a two-dimensional convolutional neural network is used to extract... The deep features,

[0065]

[0066] in, It is processed by a two-dimensional convolutional neural network. The feature output is W, where W is the convolution weight, B is the bias weight, and σ(·) is the activation function, which is ReLU.

[0067] Finally, the deep features of the enhanced difference image will be... By feeding it into a fully connected layer, we obtain the enhanced difference feature F based on the channel attention mechanism. DI .

[0068] Step 4: Extract the difference depth features F DF and enhance the difference feature F DI By directly adding them together, the fused feature FF is obtained;

[0069] In one implementation, to obtain a more comprehensive and richer feature representation, the features F extracted from the two fully connected layers are used here. DF and F DI An addition operation is performed to fuse their individual information. This step aims to combine the features extracted from the two branches to form the final features extracted by the entire network, which is here named the fused feature FF:

[0070] FF = F DF +F DI ;

[0071] Step 5: Construct F based on Kullback-Leibler divergence. DF and F DI of The loss function promotes feature fusion, and the fused features FF are input into the Softmax layer to obtain the change detection results.

[0072] In one implementation, the Siamese network and the convolutional neural network constitute a hyperspectral image domain adaptive change detection network; the hyperspectral image domain adaptive change detection network is a trained network, and the loss function used during training is... for:

[0073]

[0074] in, The cross-entropy loss function for binary classification is... Let the relevance alignment loss function be for domain adaptation. Y represents the Kullback-Leibler divergence loss function for feature fusion; x represents the specific value of each feature; n represents the number of samples; and Y represents the number of samples. i and Represent the true label and predicted label of sample i, respectively; λ1 and λ2 are... and Hyperparameters.

[0075] This embodiment also provides a hyperspectral image domain adaptation change detection device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the hyperspectral image domain adaptation change detection method.

[0076] This embodiment also provides a software product, including a computer program, which, when executed by a processor, implements the hyperspectral image domain adaptive change detection method.

[0077] This embodiment also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the hyperspectral image domain adaptive change detection method.

[0078] The invention will be further illustrated by the following experiments.

[0079] This experiment was written in Python and implemented using the classic deep learning framework PyTORCH. Based on Python's remote sensing image read / write functions, it aims to discover variation information between two hyperspectral images of size C×H×W from different time points, where C, H, and W represent the spectral dimension, length, and width of the hyperspectral image, respectively. The Python remote sensing image read / write functions are publicly available technology in this field and will not be described in detail here.

[0080] In this experiment, two hyperspectral datasets named "China" and "Rivers" were selected as datasets to verify the effectiveness of the model and to validate the accuracy and reliability of the invention. The "China" dataset was acquired on May 3, 2006, and April 23, 2007, over farmland in Yancheng City, Jiangsu Province, China. This dataset was captured using the Hyperion sensor of the EO-1 satellite, with a wavelength range of 0.4 to 2.5 micrometers, a spatial resolution of approximately 30 meters, and a spectral resolution of approximately 10 nanometers. Each image has 420×140 pixels and includes 154 spectral bands. The changes observed in this dataset are primarily due to crop rotation. The "Rivers" dataset was acquired on May 3, 2013, and December 31, 2013, in a river region of Jiangsu Province, China. This dataset was also captured using the Hyperion sensor of the EO-1 satellite, with a wavelength range of 0.4 to 2.5 micrometers, a spatial resolution of approximately 30 meters, and a spectral resolution of approximately 10 nanometers. Each image consists of 463×241 pixels and includes 198 spectral bands. The changes observed in this dataset are mainly due to the disappearance of substances in the river. To comprehensively evaluate the performance of the method of this invention, we compare it with the method of this invention using the following methods: classical differential principal component analysis (Method 1), support vector machine (Method 2), deep learning multilayer encoder-decoder attention network (Method 3), self-supervised hyperspectral spatial-spectral feature understanding network (Method 4), two-branch differential magnification convolution (Method 5), and multi-scale differential change feature fusion network (Method 6).

[0081] Experimental setup: For both datasets, 5% of the samples were used for training and 95% for testing. The China dataset used 3150 samples for training and 59850 samples for testing, while the river dataset used 5579 samples for training and 106004 samples for testing.

[0082] Evaluation metrics: Overall Accuracy (OA), Kappa coefficient, Precision, Recall, and F1 score (F1). These metrics are commonly used quantitative evaluation metrics in hyperspectral image change detection and can be calculated using a confusion matrix.

[0083] Table 1 Confusion Matrix

[0084]

[0085]

[0086] In Table 1, TP represents the number of pixels in the true label that changed and were correctly predicted as changed. FP represents the number of pixels in the true label that did not change and were incorrectly predicted as changed. FN represents the number of pixels in the true label that changed and were incorrectly predicted as unchanged. TN represents the number of pixels in the true label that did not change and were correctly predicted as unchanged.

[0087] The calculation formulas for these five evaluation indicators are as follows:

[0088]

[0089] Table 2 Comparative test results

[0090]

[0091] As can be clearly observed from the data comparison in Table 2, the method of this invention exhibits superior performance compared to other methods. This invention achieves higher values ​​in all three key metrics: overall accuracy, Kappa coefficient, and F1 score. The improved Kappa coefficient, a commonly used indicator for measuring classification accuracy, signifies that this invention demonstrates higher accuracy and stability in distinguishing between changed and unchanged data. The improved F1 score further demonstrates the excellent performance of this invention in balancing accuracy and recall. Therefore, the evaluation of the Kappa coefficient and F1 score is more objective and comprehensive.

[0092] Compared to traditional change detection methods, the method of this invention achieves a better balance between accuracy and recall. Compared with change detection methods of similar depth, this invention demonstrates a significant advantage in improving the ability to recognize change information. In summary, the method of this invention exhibits significant advantages and unique performance in change detection. It achieves high scores in overall accuracy, Kappa coefficient, and F1 score. This invention aims to improve the performance of change detection in hyperspectral images, providing strong support for the research and application of change detection.

[0093] The key point of this invention is the proposal of a hyperspectral image domain-adaptive change detection method that integrates correlation alignment and spectral attention features. This method accomplishes the change detection task based on domain-adaptive correlation alignment, channel attention mechanisms, and Kullback-Leibler divergence. Based on a convolutional neural network, this invention reduces the difference in feature distribution between images at different times through correlation alignment, thus reducing the network's ability to over-predict changes. Furthermore, this invention increases the correlation of the channel dimensions of the difference images through an attention mechanism and further mines semantic information, improving the network's ability to recognize subtle changes. Finally, this invention employs Kullback-Leibler divergence loss to fuse the features extracted from the two branches, thereby further enhancing the network's accuracy in change detection.

[0094] It should be understood that the embodiments described above are only some, not all, of the embodiments of the present invention. Furthermore, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form feasible technical solutions. Such combinations are not constrained by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0095] It should be understood that the above description of the preferred embodiments is quite detailed, but it should not be considered as a limitation on the scope of protection of this invention. Those skilled in the art, under the guidance of this invention, can make substitutions or modifications without departing from the scope of protection of the claims of this invention, and all such substitutions or modifications fall within the scope of protection of this invention. The scope of protection of this invention should be determined by the appended claims.

Claims

1. A method for detecting hyperspectral image domain adaptation changes, characterized in that, Includes the following steps: Step 1: Obtain hyperspectral remote sensing images of the same region at the first time point. and hyperspectral remote sensing images acquired at the second time point Where C, H, and W refer to the spectral dimension, length, and width of the hyperspectral remote sensing image, respectively, and the depth features T1 and T2 are extracted based on the Siamese network. fea and T2 fea And construct differential deep features F through deep features DF ; Step 2: Extract deep features T1 and T2 based on Siamese network. fea and T2 fea Furthermore, a correlation alignment loss is constructed based on deep features. Step 3: Extract the enhanced difference feature F between T1 and T2 using a convolutional neural network based on the channel attention mechanism. DI ; Step 4: Extract the difference depth features F DF and enhance the difference feature F DI By directly adding them together, the fused feature FF is obtained; Step 5: Construct F based on Kullback-Leibler divergence. DF and F DI loss function The feature fusion is promoted, and the fused feature FF is input into the Softmax layer to obtain the change detection result.

2. The hyperspectral image domain adaptation change detection method according to claim 1, characterized in that: In step 1, the Siamese network first processes T1 or T2 through a two-dimensional convolutional neural network and then outputs the results. in, It is the feature output of phase T1 or T2 after processing by a two-dimensional convolutional neural network, W 1 / 2 For convolution weights, B 1 / 2 The bias weights are σ(·), and the activation function is σ(·). Then... and Subtract them and take the absolute value to obtain the difference features. Where |·| represents the absolute value operation; then the difference features are input into a two-dimensional convolutional neural network to extract deep difference features, and finally the deep difference features T are... DF The differential depth features F are obtained by feeding them into a fully connected layer. DF .

3. The hyperspectral image domain adaptation change detection method according to claim 1, characterized in that: In step 2, the correlation alignment loss is based on the Siamese network to extract the deep features T1 and T2. fea and T2 fea ; During training, one of the loss functions used is the domain-adaptive relevance alignment loss function. in, Let n1 represent the square of the Frobenius criterion matrix, 1 represent a column vector consisting entirely of 1s, and n1 and n2 represent the number of samples in the two time phases T1 and T2, respectively.

4. The hyperspectral image domain adaptation change detection method according to claim 1, characterized in that: In step 3, enhanced difference features F are extracted based on a convolutional neural network. DI The network includes a channel attention mechanism, a two-dimensional convolutional neural network, and fully connected layers. The channel attention mechanism directly subtracts the original images of two time phases, T1 and T2, and performs an absolute value operation to obtain the difference image T. DI Then, the difference image T is enhanced based on the channel attention mechanism. DI Dimensional features, to obtain enhanced features Then, a two-dimensional convolutional neural network is used to extract... deep features Finally, the deep features of the difference image will be enhanced. By feeding it into a fully connected layer, we obtain an enhanced difference feature F based on a channel attention mechanism. DI ; in, Depend on Reconstructed; Depend on and Obtained by matrix multiplication; It is a scaling parameter; in, It is processed by a 2D convolutional neural network. The feature output is W, where W is the convolution weight, B is the bias weight, and σ(·) is the activation function.

5. The hyperspectral image domain adaptation change detection method according to any one of claims 1-4, characterized in that: The Siamese network and convolutional neural network constitute a hyperspectral image domain adaptive change detection network; the hyperspectral image domain adaptive change detection network is a pre-trained network, and the loss function used during training is... for: in, The cross-entropy loss function for binary classification. Let be the relevance alignment loss function for domain adaptation. Y represents the Kullback-Leibler divergence loss function for feature fusion; x represents the specific value of each feature; n represents the number of samples; and Y represents the number of samples. i and λ1 and λ2 represent the true label and predicted label of sample i, respectively; λ1 and λ2 are... and Hyperparameters.

6. A hyperspectral image domain adaptation change detection device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the hyperspectral image domain adaptive change detection method as described in any one of claims 1 to 5.

7. A software product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the hyperspectral image domain adaptive change detection method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the hyperspectral image domain adaptive change detection method as described in any one of claims 1 to 5.

Images