Land change detection system based on generative adversarial network

Abstract

The invention relates to the technical field of neural networks, and further relates to a land change detection system based on a generative adversarial network. The multi-source data fusion and feature extraction part is used for performing high-dimensional fusion on the multi-source remote sensing data, considering a time sequence and spatial distribution, ensuring time-space consistency and integrity of information, and obtaining fused data; introducing a space-time attention mechanism, and adaptively adjusting the importance of pyramid features to obtain an attention weight; the generative adversarial network part is used for designing a conditional generative adversarial network and obtaining conditional features based on pyramid features and attention weights; and the land change detection part is used for taking gradient change and Euclidean distance of the generated image into consideration, and performing change detection based on the generated image pair to obtain a change detection score. The invention provides an efficient and flexible solution for land change monitoring and environmental protection.

Description

Technical Field

[0001] The present invention belongs to the technical field of neural network, and in particular relates to a land change detection system based on a generative adversarial network. Background Art

[0002] With the rapid development of remote sensing technology, geographic information systems (GIS) and remote sensing image analysis have been widely used in the fields of land change monitoring, environmental protection, urban planning and resource management. Land change detection is a method that uses multi-temporal remote sensing data to identify and quantify surface changes. It extracts surface change information by analyzing image data at different time points, which can help governments and enterprises better understand land use, monitor the impact of natural disasters and manage the ecological environment. However, with the explosive growth of remote sensing data and the complexity of surface changes, traditional land change detection technology seems to be unable to cope with large-scale data and complex surface features.

[0003] At present, the mainstream methods of land change detection are mainly divided into two categories: pixel-based change detection methods and object-based change detection methods. Pixel-based change detection methods are one of the most traditional change detection methods. It identifies surface changes by directly comparing the changes in pixel values ​​at the same location in multi-temporal remote sensing images. Common pixel-level change detection algorithms include image difference method, image ratio method, change vector analysis (CVA) and principal component analysis (PCA). The advantages of these methods are simple calculation and convenient implementation, and they are suitable for areas with obvious surface changes. However, the disadvantages of this type of method are also very obvious: pixel-level detection has high requirements for geometric correction and spectral consistency of images, and is easily affected by factors such as image noise, shadows and seasonal changes, resulting in inaccurate detection results. In addition, pixel-level methods cannot make full use of the spatial information of images, and are prone to "salt and pepper effect", that is, there are a large number of noise points in the detection results. In order to overcome the shortcomings of pixel-level methods, object-based change detection methods have emerged. This method first segments the remote sensing image, divides the image into multiple physically meaningful objects or regions, and then performs change detection at the object level. Object-level methods can make full use of the spectral, spatial and texture information of images, reduce the impact of noise and improve detection accuracy. Common object-level methods include object-oriented image analysis (OBIA) and multi-scale segmentation methods. However, a major problem with object-level methods is that the segmentation accuracy depends on the resolution of the image and the parameter settings of the algorithm. Different parameter settings may lead to completely different segmentation results. In addition, object-level methods still have problems of insufficient detection accuracy and low computational efficiency when facing complex surfaces and multi-scale changes. Summary of the invention

[0004] The main purpose of the present invention is to provide a land change detection system based on a generative adversarial network, which can accurately extract the multi-scale features of surface changes, generate images that are highly similar to real changes, and evaluate the degree of surface changes through comprehensive change detection scores. The system significantly improves the accuracy and robustness of change detection, adapts to a variety of complex surface environments and data types, and solves the problems of insufficient multi-source data fusion, poor sensitivity to non-stationary spatiotemporal features, and insufficient multi-scale change detection capabilities in the prior art, providing an efficient and flexible solution for land change monitoring and environmental protection.

[0005] In order to solve the above problems, the technical solution of the present invention is achieved as follows:

[0006] A land change detection system based on a generative adversarial network, the system comprising: a multi-source data fusion and feature extraction part, used for high-dimensional fusion of multi-source remote sensing data, considering time series and spatial distribution, ensuring the spatiotemporal consistency and integrity of information, and obtaining fused data; an adaptive convolution kernel design is used to capture non-stationary spatiotemporal features in the fused data; based on the non-stationary spatiotemporal features, a multi-scale feature pyramid construction method is used to ensure accurate detection of changes at different scales, and pyramid features at different scales are obtained through downsampling and upsampling operations of multi-layer features; a spatiotemporal attention mechanism is introduced to adaptively adjust the importance of pyramid features and obtain attention weights; a generative adversarial network part, used for designing a conditional generative adversarial network, obtaining conditional features based on pyramid features and attention weights, a generator generates images after simulated changes, and a discriminator guides the improvement of the generator by evaluating the authenticity of these images; a land change detection part, used for considering the gradient change and Euclidean distance of the generated image, performing change detection based on the generated image pairs, and obtaining a change detection score.

[0007] Furthermore, in the multi-source data fusion and feature extraction part, multi-source remote sensing data is fused in high dimensions, taking into account time series and spatial distribution to ensure the temporal and spatial consistency and integrity of information. The process of obtaining fused data specifically includes: collecting raw data from n different remote sensing data sources; using interpolation and resampling to align all raw data in time and space to ensure that all raw data are on the same spatiotemporal grid, with equal weight for each data source, and the obtained fused data is expressed using the following formula:

[0008]

[0009] in, is the fused data set, which is a function of time t and spatial position (x, y); x is the X-axis coordinate; y is the Y-axis coordinate; T is the total time span; Ω is the spatial domain; is the signal strength of the ith data source, τ is the time offset, u is the X-axis offset, v is the Y-axis offset, λ i is the time decay coefficient of the i-th data source; σ i is the spatial distribution standard deviation of the ith data source; Δt i is the temporal resolution of the ith data source; P i is the periodic change period of the i-th data source; α i is the periodic variation intensity coefficient of the i-th data source; is a Gaussian kernel used to simulate spatial correlation, σ i is the correlation range parameter, which controls the range of the control correlation; Used to handle differences in temporal resolution between different data sources.

[0010] Furthermore, in the multi-source data fusion and feature extraction part, adaptive convolution kernel design is used to capture the non-stationary spatiotemporal features F in the fused data through the following formula:

[0011]

[0012] Among them, Z is the normalization constant, which ensures that the integral of the convolution kernel is 1; σ t is the preset time dimension adaptive standard deviation; σ s Adaptive standard deviation for preset spatial dimensions.

[0013] Furthermore, in the multi-source data fusion and feature extraction part, pyramid features of different scales are obtained through the following formula:

[0014]

[0015] Among them, P l is the pyramid feature corresponding to the lth level scale; DownSample l represents the lth downsampling operation; is the Laplace operator; represents the kth upsampling operation; P k is the pyramid feature corresponding to the k-th level scale.

[0016] Furthermore, in the multi-source data fusion and feature extraction part, the spatiotemporal attention mechanism is introduced to adaptively adjust the importance of pyramid features. The specific process of obtaining the attention weight includes: calculating the spatiotemporal correlation matrix according to the features of each layer of the pyramid through the following formula:

[0017]

[0018] Among them, C ij is the element in the i-th row and j-th column of the spatiotemporal correlation matrix; t i, t j are the time of feature i and feature j respectively; the attention weight of each layer of pyramid features is calculated using the spatiotemporal correlation matrix:

[0019]

[0020] Among them, A l is the attention weight of the l-th level scale pyramid feature at time t and spatial position (x, y); F i , F j is the feature of feature i and feature j at time t and spatial position (x, y); σ F is the importance parameter for controlling the similarity of eigenvalues ​​and is the set value.

[0021] Furthermore, the formula of the generator of the conditional generative adversarial network designed in the generative adversarial network part is expressed as follows:

[0022]

[0023] Among them, z is a random noise vector, which follows a normal distribution. P′ l is the conditional feature of the lth layer, which is calculated by the following formula:

[0024] P l ′=A l ·P l ;

[0025] A l is the attention weight of the l-th level scale pyramid feature; G(z, P′, A) is the simulated changed image generated by the generator; W G and b G are the weight matrix and bias vector of the generator respectively; φ(·) is the nonlinear activation function, which is the ReLU function or the LeakyReLU function; σ(·) is the output layer activation function, which is the tanh function or the sigmoid function; U is the weight matrix of the conditional input of the generator, which is used to fuse the conditional features and noise; A is the attention weight matrix; z is the noise; the output of the generator is a simulated image, which is as similar as possible to the real image in high-dimensional space.

[0026] Furthermore, the formula of the discriminator is expressed as follows:

[0027]

[0028] Among them, I is the input image, which is the real image I real Or the simulated image G(z, P′, A) generated by the generator; D(I, P′, A) is the probability output by the discriminator, indicating the authenticity of the input image I; W Dand b D are the weight matrix and bias vector of the discriminator respectively; V is the weight matrix of the conditional input of the discriminator; D(I, P′, A) is the probability of the discriminator output, indicating the authenticity of the input image I.

[0029] Furthermore, the loss function of the generator is As shown below:

[0030]

[0031] Among them, p z (z) is the probability distribution of random noise z; is the expected operator; P′ is the conditional feature set; is the basic generator loss, which is used to maximize the probability that the generated image is recognized as a real image by the discriminator; is a gradient normalization regularization term used to balance the learning speed of the generator on different conditional features; and are the gradients of the generator and conditional features, α k is the adaptive weight of each feature layer, ∈ is a constant to prevent division by zero; It is a reconstruction loss term, which ensures the consistency between the generated image and the real image in the conditional feature space. The L2 norm is used to measure the difference between the generated image and the real image; ||||2 represents the L2 norm operation; represents the loss part corresponding to the conditional features of the kth layer; |||| represents the operation of finding the determinant of the matrix or the modulus of the vector; p P′ (P′) is the probability distribution of conditional feature P′; I real (P′ l ) is the conditional feature of the real image at the lth level scale;

[0032] Furthermore, the loss function of the discriminator is As shown below:.

[0033]

[0034] Among them, p data (I real ) is the probability distribution of the real image; 1-D(G(z, P′, A), P′, A) represents the authenticity judgment output of the discriminator on the generated image, and the probability that the generated image is false is obtained;

[0035] is the first discriminator loss, which is used to maximize the probability that the discriminator correctly identifies the real image; ] is the second discriminator loss, which is used to maximize the probability that the discriminator correctly identifies the generated image; is a multi-scale contrast loss term that enhances the sensitivity of the discriminator to subtle changes by using the difference in the discriminator outputs at different scales; p and q are both the norms of the contrast amount.

[0036] Furthermore, in the land change detection part, the change detection is performed based on the generated image pairs to obtain the change detection score through the following formula:

[0037]

[0038] Where, is the final change detection score, which is used to quantify the changes between the two generated images. The higher the score, the more significant the change. N represents the total number of pixels. G1(x i ,y i ) and G2(x i ,y i ): respectively represent different times at the position (x i ,y i ) the pixel values ​​of the first and second generated images; and Respectively represent the first and second generated images at position (x i ,y i ); γ: a weight parameter with a value range of [0, 1], which is used to balance the importance of gradient change and Euclidean distance in change detection.

[0039] The land change detection system based on generative adversarial network of the present invention has the following beneficial effects:

[0040] First of all, the present invention has significant advantages in multi-source data fusion and feature extraction. Traditional land change detection methods often rely on a single data source or simple data fusion technology, and cannot fully utilize the diversity and richness of multi-source remote sensing data. By introducing an adaptive convolution kernel design, the present invention can dynamically adjust the shape and size of the convolution kernel, so as to better capture the non-stationary spatiotemporal features in the fused data. The use of adaptive convolution kernels enables the system to extract more delicate change features at different time and space scales, enhancing the system's sensitivity and adaptability to complex surface changes. This adaptive feature extraction method not only improves the efficiency of data utilization, but also significantly improves the accuracy of change detection, especially in the case of multi-source data fusion, it can give full play to the advantages of different data sources.

[0041] Secondly, the present invention adopts a spatiotemporal attention mechanism to adaptively adjust the importance of pyramid features. The introduction of this mechanism effectively solves the problem of inflexible weight allocation for different change areas in the prior art. By calculating the spatiotemporal correlation matrix of each layer of pyramid features, the system can adaptively adjust the weight of each feature in change detection according to the temporal and spatial correlation of the features. This attention mechanism ensures that the system can pay more attention to areas with significant changes in change detection, while reducing attention to unimportant areas, thereby improving the accuracy and efficiency of detection. Especially when processing large-scale remote sensing data and complex surface environments, the introduction of the spatiotemporal attention mechanism enables the system to dynamically optimize the allocation of feature weights and ensure accurate identification and detection of key change areas.

[0042] In addition, the multi-scale feature pyramid construction method of the present invention further enhances the system's ability to detect changes at different scales. Surface changes often manifest as multi-scale, multi-level feature changes, and traditional methods are difficult to perform comprehensive change detection at multiple scales. The present invention constructs a multi-scale feature pyramid through downsampling and upsampling operations of multiple layers of features, so that the system can effectively capture change features at different scales. In particular, by combining the Laplace operator for edge enhancement, the system performs well in capturing both details and global features. This method of constructing a multi-scale feature pyramid enables the system to achieve significant detection effects when detecting large-scale land use changes and small-scale local terrain changes, and improves the adaptability and reliability of the system in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A schematic diagram of the system structure of a land change detection system based on a generative adversarial network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0045] Example 1: Reference Figure 1, a land change detection system based on generative adversarial networks, the system includes: a multi-source data fusion and feature extraction part, which is used to perform high-dimensional fusion of multi-source remote sensing data, consider time series and spatial distribution, ensure the spatiotemporal consistency and integrity of information, and obtain fused data; an adaptive convolution kernel design is used to capture non-stationary spatiotemporal features in fused data; based on non-stationary spatiotemporal features, a multi-scale feature pyramid construction method is used to ensure accurate detection of changes at different scales, and pyramid features at different scales are obtained through downsampling and upsampling operations of multi-layer features; a spatiotemporal attention mechanism is introduced to adaptively adjust the importance of pyramid features and obtain attention weights; a generative adversarial network part is used to design a conditional generative adversarial network, and conditional features are obtained based on pyramid features and attention weights. The generator generates images after simulated changes, and the discriminator guides the improvement of the generator by evaluating the authenticity of these images; a land change detection part is used to consider the gradient change and Euclidean distance of the generated image, and change detection is performed based on the generated image pairs to obtain a change detection score.

[0046] Specifically, the present invention relates to a land change detection system based on a generative adversarial network (GAN), the core of which is to accurately detect land changes by fusing multi-source remote sensing data, using adaptive convolution kernels to capture non-stationary spatiotemporal features, constructing a multi-scale feature pyramid, introducing a spatiotemporal attention mechanism, and designing a conditional generative adversarial network. The system first starts with the multi-source data fusion and feature extraction part, the purpose of which is to perform high-dimensional fusion of information from different remote sensing data sources to ensure the consistency and integrity of the data in time series and spatial distribution. This multi-source fusion technology integrates data from different sensors and time points to make up for the shortcomings of a single data source in terms of spatial resolution, temporal resolution, and spectral characteristics, so that the system can capture information about surface changes more comprehensively and accurately. In actual operation, multi-source data fusion not only considers the spatial feature distribution, but also fully considers the time series features. By combining the change patterns of the two dimensions of time and space, the fused data can better reflect the real situation of surface changes. After the multi-source data is fused, the system enters the adaptive convolution kernel design stage. The key to this step is to capture the non-stationary spatiotemporal features in the fused data. When processing remote sensing data, traditional convolutional neural networks usually assume that the characteristics of the data are stable. However, in reality, remote sensing data often has complex non-stationary characteristics, which reflects the changes and inconsistencies of the data at different time and space scales. In order to solve this problem, the present invention introduces an adaptive convolution kernel design. By dynamically adjusting the size and shape of the convolution kernel, the system can better adapt to the non-stationarity of the data and capture more refined spatiotemporal features. For example, in areas with significant changes, the convolution kernel may need a smaller and more concentrated shape to capture the details of the changes; while in areas with gentle changes, the convolution kernel can be larger to obtain a wider range of contextual information. Such a design enables the system to more accurately capture the characteristics of land changes when processing complex and ever-changing remote sensing data.

[0047] After obtaining non-stationary spatiotemporal features, the system adopts a multi-scale feature pyramid construction method, which is introduced to ensure that the system can accurately detect land changes at different scales. The types and scales of land changes often vary. Some changes may be only small-scale subtle adjustments, such as seasonal changes in vegetation; while some changes may be large-scale, such as large-scale land development or surface damage caused by natural disasters. The multi-scale feature pyramid can effectively capture and express these changes at different scales by combining downsampling and upsampling operations of multiple layers of features. Downsampling operations help extract low-frequency features and are suitable for large-scale land change detection, while upsampling operations help restore high-frequency features and are suitable for small-scale change detection. Through this multi-scale feature extraction method, the system can process large-scale and small-scale changes at the same time under the same network framework, ensuring the comprehensiveness and accuracy of detection. In order to further improve the flexibility and accuracy of feature selection, the system introduces a spatiotemporal attention mechanism. The main function of this mechanism is to adaptively adjust the importance of pyramid features so that the system can focus more on those areas or time periods that are most critical for land change detection. The attention mechanism performs weighted processing on different features by calculating the importance weight of each feature in the current detection task. This means that when the change features in a certain area or time period are particularly significant, the system can automatically increase the feature weight of that area or time period, making the detection results more accurate. For example, in images after certain natural disasters such as earthquakes or floods, the attention mechanism can help the system better focus on severely affected areas and ignore areas with little change. This adaptive attention mechanism effectively improves the detection efficiency and accuracy of the system, especially when processing large-scale remote sensing data.

[0048] After the above feature extraction and selection process, the system enters the generative adversarial network part. The core idea of ​​the generative adversarial network is to make the images generated by the generator closer and closer to the real change images through adversarial training, while the discriminator guides the improvement of the generator by continuously evaluating the authenticity of the generated images. In the present invention, the conditional generative adversarial network (CGAN) is designed for the land change detection task. The input of the generator is the multi-scale pyramid features and attention weights extracted through the above steps. These features and weight information provide the generator with rich contextual information about land changes. The generator generates images after simulated changes based on these conditional features, and the discriminator compares these generated images with the real images, and gradually optimizes the performance of the generator by feeding back the generation error of the generator. In this way, through the adversarial training between the generator and the discriminator, the system can generate more realistic and close to real change images, which provides a more reliable basis for subsequent change detection. Finally, the system enters the land change detection part, the main task of which is to perform change detection based on the generated image pairs. In this process, the system not only considers the gradient changes of the generated images, but also combines the Euclidean distance for change detection. Gradient changes are mainly used to capture changes in image details, which is particularly effective for identifying small features such as edge changes; while Euclidean distance is used to measure the overall difference between two images and can reflect changes on a larger scale. By combining these two methods, the system can assess land changes more comprehensively and accurately. The final change detection score is based on the gradient change and Euclidean distance calculation results of the image pair, combined with a comprehensive scoring mechanism, which can provide a quantitative assessment of the degree of change and provide a scientific basis for land change management and decision-making.

[0049] Embodiment 2: Multi-source data fusion and feature extraction part, high-dimensional fusion of multi-source remote sensing data, considering time series and spatial distribution, ensuring the temporal and spatial consistency and integrity of information, and the process of obtaining fused data specifically includes: collecting raw data from n different remote sensing data sources; using interpolation and resampling to align all raw data in time and space to ensure that all raw data are on the same spatiotemporal grid, with equal weight for each data source, and the obtained fused data is expressed using the following formula:

[0050]

[0051] in, is the fused data set, which is a function of time t and spatial position (x, y); x is the X-axis coordinate; y is the Y-axis coordinate; T is the total time span; Ω is the spatial domain; is the signal strength of the ith data source, τ is the time offset, u is the X-axis offset, v is the Y-axis offset, λ iis the time decay coefficient of the i-th data source; σ i is the spatial distribution standard deviation of the ith data source; Δt i is the temporal resolution of the ith data source; P i is the periodic change period of the i-th data source; α i is the periodic variation intensity coefficient of the i-th data source; is a Gaussian kernel used to simulate spatial correlation, σ i is the correlation range parameter, which controls the range of the control correlation; Used to handle differences in temporal resolution between different data sources.

[0052] Specifically, this fusion process first collects raw data from n different remote sensing data sources, which may include satellite images, drone images, and ground radar data. These data often have large differences due to different acquisition times, spatial resolutions, and spectral characteristics. Therefore, before fusing these data, they need to be preprocessed. In the formula, all raw data are aligned to a unified time and space grid through interpolation and resampling. This operation ensures that the information of all data sources can be represented and utilized at the same time point and spatial position. After the data is aligned, the formula further introduces a time attenuation factor This factor is used to simulate the decay characteristics of the information of each data source over time. i represents the time decay coefficient of the ith data source, reflecting the time validity of the information of the data source. As time goes by, the representativeness and accuracy of remote sensing data may gradually decline. Therefore, by introducing the time decay factor, a higher weight can be given to newer data in the fusion process, thereby improving the timeliness and accuracy of the fusion result. At the same time, in order to deal with the spatial distribution characteristics of the data, a Gaussian kernel function is introduced in the formula The Gaussian kernel function is a common smoothing function used to model spatial correlation. It is expressed by a standard deviation parameter σ i To control the spatial impact range of the data. In remote sensing data processing, different data sources have different spatial resolutions and accuracies. The introduction of Gaussian kernel function can effectively smooth and integrate data from different sources, ensuring that they can reasonably reflect the actual surface characteristics during the fusion process. By controlling the parameter σ i , the influence range of each data source in the fused data can be adjusted to ensure the spatial consistency of the fusion result. In addition, remote sensing data often have different temporal resolutions, which leads to inconsistent sampling frequencies of different data sources in time. In order to effectively handle this difference, the formula is used Function. The Sinc function is a commonly used tool in signal processing and data analysis, which is used to keep the bandwidth of the signal unchanged during the interpolation process. In the context of the present invention, the use of the sinc function can effectively handle the differences in time resolution between different data sources, ensuring that the fusion results in the time dimension are not distorted or biased. Finally, the formula also takes into account the periodic variation characteristics of the data source, through a periodic modulation term The periodic changes in remote sensing data are usually caused by natural phenomena (such as seasonal changes, day and night changes) or human activities (such as regular land cultivation). By modeling these periodic changes, the system can more accurately identify and separate the differences between signal changes caused by periodic changes and actual land change signals. i and P i They represent the periodic change intensity and period of the ith data source respectively. By adjusting these parameters, the system can more flexibly adapt to the characteristics of different data sources and improve the detection accuracy of land changes.

[0053] Embodiment 3: In the multi-source data fusion and feature extraction part, the following formula is used to design an adaptive convolution kernel to capture the non-stationary spatiotemporal features F in the fused data:

[0054]

[0055] Among them, Z is the normalization constant, which ensures that the integral of the convolution kernel is 1; σ t is the preset time dimension adaptive standard deviation; σ s Adaptive standard deviation for preset spatial dimensions.

[0056] Specifically, the formula extracts spatiotemporal features by performing triple integral operations on the fused data and combining them with the form of Gaussian kernel function. Here, the core of the convolution kernel design is to perform weighted processing through Gaussian distribution in both time and space dimensions, so that the convolution operation can not only capture local changes in the data, but also adapt to feature changes in different regions and time periods. This design uses the time dimension adaptive standard deviation σ t and spatial dimension adaptive standard deviation σ s To adjust the shape and range of the convolution kernel, it can provide different sensitivities at different time points and spatial locations. Specifically, the form of the Gaussian kernel function is It reflects the weighting strategy of data in time and space, where τ, u and v represent the time and space offsets respectively. This adaptive convolution kernel design effectively solves the limitations of convolution operations in traditional methods. Traditional convolution kernels are usually fixed to a certain size and shape, which is incapable of processing non-stationary features because it cannot adapt to the changes of data at different locations and times. The present invention introduces an adaptive standard deviation σt and σ s , so that the size and shape of the convolution kernel can be dynamically adjusted according to the changes in the actual data. For example, in the time dimension, σ t The size of the "time window" of the convolution kernel can be adjusted according to how fast the data changes; similarly, in the spatial dimension, σ s The "spatial range" of the convolution kernel can be adjusted according to the spatial distribution of the data features. This adaptive mechanism makes the convolution operation more flexible and adaptable, so that the non-stationary features in the data can be captured more accurately. In addition, in order to ensure that the weighted calculation results of the convolution kernel can reasonably reflect the distribution of the real data, the normalization constant Z is introduced into the formula. The function of this constant is to ensure that the integral result of the convolution kernel is 1, thereby avoiding data distortion caused by the non-normalization of the kernel function. The normalization process ensures that when the data is convolved, the strength of the output feature depends only on the characteristics of the data itself, and will not produce unnecessary amplification or reduction effects due to changes in the scale of the convolution kernel. This design is particularly important in the framework of generative adversarial networks, because generative adversarial networks need to be iteratively optimized between the generator and the discriminator, and any unreasonable feature extraction will directly affect the accuracy of the final change detection results. Through this adaptive convolution kernel design, the system can capture the non-stationary spatiotemporal features after the fusion of multi-source data, thereby providing more accurate and detailed feature input for the generator of the generative adversarial network. In the framework of the generative adversarial network, the generator uses these extracted spatiotemporal features to simulate images of land changes, while the discriminator guides the optimization of the generator by evaluating the authenticity of these simulated images. The adversarial training process of the generator and the discriminator depends on the quality and expressiveness of the input features. Therefore, the effectiveness of the adaptive convolution kernel in the feature extraction stage directly determines the overall performance of the generative adversarial network. The adaptive convolution kernel design not only improves the flexibility and accuracy of feature extraction, but also enhances the robustness of the system in dealing with complex surface changes, enabling the land change detection system to more accurately identify and evaluate various types of land changes.

[0057] Embodiment 4: In the multi-source data fusion and feature extraction part, pyramid features of different scales are obtained through the following formula:

[0058]

[0059] Among them, P l is the pyramid feature corresponding to the lth level scale; DownSample l represents the lth downsampling operation; is the Laplace operator; represents the kth upsampling operation; P k is the pyramid feature corresponding to the k-th level scale.

[0060] Specifically, since remote sensing data is essentially a multi-scale and multi-level observation of the earth's surface, different types of surface changes (such as vegetation coverage, building changes, soil erosion, etc.) may be more significant or hidden at different scales. Therefore, single-scale feature extraction methods often cannot fully reflect complex surface changes, and the introduction of multi-scale feature pyramids provides an effective way to solve this problem. In the present invention, the construction of the feature pyramid is achieved by performing multiple downsampling (DownSample) and upsampling operations on the feature map. The downsampling operation aims to reduce the resolution of the feature map to capture large-scale surface changes; the upsampling operation is to restore the detail information in the lower-resolution feature map, so that the system can pay attention to small-scale change details at the same time. In this way, the system obtains multi-level feature representations at different scales, and each layer of feature maps represents a different spatial scale. In order to further enrich and refine these feature representations, the Laplace operator is introduced into the formula Application. The Laplacian operator is a second-order differential operator, which is often used in image processing to enhance edge features. It can effectively capture the changing areas and edge information in the image. In the process of constructing the multi-scale feature pyramid, the application of the Laplacian operator can further highlight and amplify the change information in the feature map, so that the generative adversarial network can more accurately capture the change details when generating simulated change images. In addition, in order to ensure that the feature information between different levels of the multi-scale pyramid can be effectively fused, an accumulation and upsampling strategy is also adopted in the formula. Specifically, the feature map of the lth layer of the pyramid not only depends on the downsampling operation of this layer (DownSample l (F)), and also by upsampling the feature maps of the first l-1 layers (U k) and Laplace transform, and this accumulation method enables the high-resolution features of the lower level to be effectively fed back to the low-resolution features of the higher level. This bottom-up feature fusion strategy can effectively integrate feature information of different scales, ensuring that both the macro information of large-scale features and the detailed information of small-scale features are retained in each layer of the feature pyramid. In this way, the system can be more comprehensive and accurate in land change detection, especially in dealing with complex scenes of multi-scale surface changes. At the same time, this multi-scale feature pyramid construction method has unique advantages in the application of generative adversarial networks. Generative adversarial networks rely on the game between the generator and the discriminator to optimize model performance. In the context of land change detection, the generator needs to generate simulated images similar to real changes, while the discriminator evaluates the authenticity of these images and reversely optimizes the generator. In this process, the rich feature representation provided by the multi-scale feature pyramid can provide the generator with more detailed and comprehensive input information, thereby improving the quality and authenticity of the generated images. In particular, by combining the Laplacian operator with up- and down-sampling operations, the generator can more accurately reproduce the changing features of different scales when generating simulated images, making it more difficult for the discriminator to distinguish between real images and generated images, ultimately improving the performance of the entire generative adversarial network.

[0061] Embodiment 5: In the multi-source data fusion and feature extraction part, a spatiotemporal attention mechanism is introduced to adaptively adjust the importance of pyramid features. The specific process of obtaining the attention weight includes: calculating the spatiotemporal correlation matrix according to each layer of pyramid features through the following formula:

[0062]

[0063] Among them, C ij is the element in the i-th row and j-th column of the spatiotemporal correlation matrix; t i , t j are the time of feature i and feature j respectively; the attention weight of each layer of pyramid features is calculated using the spatiotemporal correlation matrix:

[0064]

[0065] Among them, A l is the attention weight of the l-th level scale pyramid feature at time t and spatial position (x, y); F i , F j is the feature of feature i and feature j at time t and spatial position (x, y); σ F is the importance parameter for controlling the similarity of eigenvalues ​​and is the set value.

[0066] Specifically, the correlation matrix quantifies the spatiotemporal relationship between different features in the form of a Gaussian kernel function, that is, considering the difference in the time dimension of the features (t i -t j ) and the difference in spatial dimensions (x i -x j ) and (y i -y j ). This design uses the characteristics of Gaussian distribution to measure similarity: when the time and space distances of two features are close, the correlation is strong, which means that the two features may have similar behaviors in a certain change pattern. Therefore, the formula attenuates the time and space differences through an exponential function, so that the correlation value can reasonably reflect the true spatiotemporal relationship between the features. This process is very critical because it provides basic data for the subsequent attention weight calculation, allowing the system to dynamically adjust according to the relative importance of the features. After obtaining the spatiotemporal correlation matrix, the system further calculates the attention weights of each layer of the pyramid features. The calculation of attention weights is based on a weighted feature similarity measure, which involves the Euclidean distance between each pair of features, that is, |F i -F j | 2 This measurement method fully considers the spatial and temporal changes of features and maps these changes into a similarity measure through an exponential function. Specifically, this method introduces a parameter σ F To control the influence range of feature similarity, σ F The larger the value, the smaller the impact of feature similarity on attention weight, and vice versa. In this way, the system can adaptively adjust the attention weight of features under different change intensities and change modes, so that those features that are more important in change detection receive more attention, while the impact of those secondary features is appropriately weakened. This adaptive adjustment mechanism of weights enables the system to make more accurate judgments at the feature extraction stage, optimizing the performance of the entire detection process.

[0067] In addition, the spatiotemporal attention mechanism of the present invention also pays special attention to the balance between local and global features of features. In the context of multi-source data fusion, different data sources may have different temporal resolutions and spatial resolutions, which requires the system to not only capture large-scale global change features, but also to be able to identify and analyze small-scale local change details. Through the above-mentioned spatiotemporal correlation matrix and attention weight calculation method, the present invention can effectively balance the contribution of global and local features, so that the generative adversarial network can maintain global consistency without losing local details when simulating and detecting surface changes. This ability is crucial for land change detection, because surface changes often have both large-scale significant changes (such as deforestation, urban expansion) and subtle local changes (such as small-scale building renovation or surface erosion). Only by achieving a balance between global and local features can the system accurately identify and locate the change area in a complex and changeable surface environment. Furthermore, the spatiotemporal attention mechanism plays a crucial role in the framework of the generative adversarial network. The generative adversarial network optimizes the detection capability of the system through the game between the generator and the discriminator, in which the generator is responsible for simulating real surface changes, while the discriminator is responsible for judging the difference between the generated image and the real image. In this adversarial training process, the generator relies on the quality of the input features to generate change images that are as close to the real ones as possible, while the discriminator optimizes the feedback according to the real degree of these images. By introducing the spatiotemporal attention mechanism, the system can provide the generator with more accurate and targeted feature information, making the generated change images more realistic and detailed. This improvement not only improves the training efficiency of the generative adversarial network, but also significantly enhances the generalization ability of the system, making it perform better when processing remote sensing data of different types and complexity. The introduction of the spatiotemporal attention mechanism also helps the generative adversarial network overcome some inherent problems in traditional methods. For example, traditional land change detection methods usually rely on fixed feature extraction and matching strategies, which are often insufficient when processing non-stationary spatiotemporal data. The present invention enables the system to dynamically adjust the detection strategy according to the actual situation of the change area by adaptively adjusting the feature weights. This flexibility greatly improves the accuracy and robustness of detection, allowing the system to better adapt to various complex situations in practical applications. Especially when facing large-scale remote sensing data, the spatiotemporal attention mechanism can significantly reduce the computational complexity because it can selectively focus on areas with significant changes, thereby reducing unnecessary computational overhead and improving overall detection efficiency.

[0068] Example 6: Generate adversarial network The generator formula of the conditional generative adversarial network designed in part is as follows:

[0069]

[0070] Among them, z is a random noise vector, which follows a normal distribution. P′ l is the conditional feature of the lth layer, which is calculated by the following formula:

[0071] P′ l =A l ·P l ;

[0072] A l is the attention weight of the l-th level scale pyramid feature; G(z, P′, A) is the simulated changed image generated by the generator; W G and b G are the weight matrix and bias vector of the generator respectively; φ(·) is the nonlinear activation function, which is the ReLU function or the LeakyReLU function; σ(·) is the output layer activation function, which is the tanh function or the sigmoid function; U is the weight matrix of the conditional input of the generator, which is used to fuse the conditional features and noise; A is the attention weight matrix; z is the noise; the output of the generator is a simulated image, which is as similar as possible to the real image in high-dimensional space.

[0073] Specifically, the input of the generator first contains a random noise vector z, which follows the standard normal distribution N(0, 1). The introduction of random noise is a common strategy in generative adversarial networks (GANs), which aims to provide initial randomness for the generation process and ensure that the generated images are diverse and unpredictable. The diversity of the noise vector can effectively prevent the generator from falling into the trap of "mode collapse", that is, the generator only learns to produce a limited set of images and lacks diversity. Combined with the noise vector is the conditional feature P′ unique to the conditional generative adversarial network. l , these features are obtained by l The weight is obtained by weighted processing, and the weight is provided by the spatiotemporal attention mechanism, that is, P′ l =A l ·P l In this way, the conditional features effectively integrate the information of multi-scale features and attention mechanisms, ensuring that the generator can better generate images based on the spatiotemporal correlation of the input data. In order to achieve the effective fusion of conditional features and noise vectors, a fusion operation is introduced in the design of the generator. This operation is completed by the weight matrix U, which is specifically expressed as the sum of the random noise vector z and the conditional features of all layers. Fusion. This fusion step ensures that the input of the generator can comprehensively consider the characteristics of randomness and input data to form an input vector with rich information. Subsequently, this fused vector is nonlinearly mapped through a nonlinear activation function φ(·), usually a ReLU or LeakyReLU function, to further enhance the expressive power of the generator. The use of nonlinear activation functions can effectively capture complex input data patterns, enabling the generator to better model the complexity of land changes in high-dimensional space. The output layer of the generator uses an activation function σ(·), which can be a tanh function or a sigmoid function. These activation functions are used to limit the output of the generator to a specific range to ensure that the generated image values ​​are physically reasonable. For example, the tanh function limits the output value to between [-1, 1], which is very effective for normalizing image data; while the sigmoid function limits the output value to between [0, 1], which is very useful for generating grayscale images or probability images. The weight matrix W of the generator G and the bias vector b G The ultimate goal of continuous optimization through adversarial training is to generate a simulated changing image that is as close to the real image as possible in the high-dimensional feature space.

[0074] Example 7: The formula of the discriminator is as follows:

[0075]

[0076] Among them, I is the input image, which is the real image I real Or the simulated image G(z, P′, A) generated by the generator; D(I, P′, A) is the probability output by the discriminator, indicating the authenticity of the input image I; W D and b D are the weight matrix and bias vector of the discriminator respectively; V is the weight matrix of the conditional input of the discriminator; D(I, P′, A) is the probability of the discriminator output, indicating the authenticity of the input image I.

[0077] Specifically, the main task of the discriminator is to distinguish whether the input image comes from real data or data simulated by the generator, so as to conduct adversarial training with the generator under the framework of generative adversarial network (GAN) to optimize the performance and detection accuracy of the system. real Or the simulated image G(z, P′, A) generated by the generator, also includes the multi-scale feature pyramid conditional feature P′ land the corresponding attention weight A. This design idea aims to ensure that the discriminator can combine multi-dimensional information to make a more comprehensive and accurate judgment when evaluating the authenticity of the image. The discriminator receives an image input I, which can be either a real image or a generated image, and also inputs a pyramid feature P′ l and the feature vector of attention weights. In the formula, the input vector is the sum of image I and all layer conditional features This combination of inputs ensures that the discriminator can consider both the visual and structural features of the image. The weight matrix V is responsible for weighted fusion of this input vector, which is then transformed nonlinearly through a nonlinear activation function φ(·) (usually ReLU or LeakyReLU). The application of nonlinear activation functions is intended to increase the discriminator's ability to capture complex feature patterns, enabling it to distinguish more subtle changes and feature differences. The result after nonlinear transformation is then combined with the discriminator's weight matrix W D Perform a dot product operation and add the bias vector b D. This linear transformation step is to adjust the feature space so that the output of the discriminator can more accurately reflect the authenticity of the input image. Finally, the discriminator maps the output value to the interval [0, 1] through the output layer activation function σ(·) (usually a sigmoid function), and the output result can be interpreted as the probability value of the image being real. The sigmoid function is chosen here because it can well represent the characteristics of the probability value, limit the output of the discriminator to a reasonable range, and make the result easier to interpret and optimize probabilistically. The core of the design of the discriminator lies in its comprehensive utilization of multi-source information. By combining the features of the input image and the pyramid conditional features, the discriminator can make a more detailed and accurate judgment on the authenticity of the image. Especially in the context of land change detection in remote sensing data, different types of surface changes will show different visual and structural features in the image, and the discriminator needs to be able to recognize the subtle differences between these features. Through the introduction of conditional features and attention mechanisms, the discriminator not only pays attention to the global features of the image, but also can deeply understand the characteristic patterns of local changes, thereby improving its ability to distinguish between real images and generated images. This design enables the generator and the discriminator to continuously promote each other during the adversarial training process. The generator will continuously optimize its generation strategy according to the feedback of the discriminator, and the discriminator will improve its discrimination ability as the generator improves. Ultimately, the adversarial learning process of this generative adversarial network can greatly improve the detection accuracy and robustness of the system for land changes. The discriminator becomes a key link in the entire system through this design based on multi-scale features and attention mechanism. It is not only the goal of generator optimization, but also one of the core components to ensure the effective training of the generative adversarial network. Through this dual input and conditional discrimination strategy, the present invention can demonstrate stronger detection capabilities and higher efficiency in complex remote sensing data analysis tasks, especially in the processing of high-dimensional and multi-scale features. In this way, the discriminator is not just a simple true or false judgment, but through deep mining and complex analysis of features, it ensures that every result output by the system has a high degree of reliability and accuracy.

[0078] Example 8: Generator loss function As shown below:

[0079]

[0080] Among them, p z (z) is the probability distribution of random noise z; is the expected operator; P′ is the conditional feature set; is the basic generator loss, which is used to maximize the probability that the generated image is recognized as a real image by the discriminator; is a gradient normalization regularization term used to balance the learning speed of the generator on different conditional features; and are the gradients of the generator and conditional features, α k is the adaptive weight of each feature layer, ∈ is a constant to prevent division by zero; It is a reconstruction loss term, which ensures the consistency between the generated image and the real image in the conditional feature space. The L2 norm is used to measure the difference between the generated image and the real image; ||||2 represents the L2 norm operation; represents the loss part corresponding to the conditional features of the kth layer; |||| represents the operation of finding the determinant of the matrix or the modulus of the vector; p P′ (P′) is the probability distribution of conditional feature P′; I real (P′ l ) is the conditional feature of the real image at the lth level scale.

[0081] Specifically, the generator’s loss function contains a basic generator loss part, which is used to maximize the probability that the generated image is recognized as a real image by the discriminator. This part of the loss is implemented through an adversarial learning mechanism, which uses the expectation operator to optimize the probability distribution of the generated image. Specifically, the generator tries to minimize the performance of the discriminator in distinguishing between real images and generated images, that is, by minimizing To maximize the probability that the discriminator identifies the generated image as a real image. In this adversarial framework, the generator continuously improves its generation strategy, making the generated images more and more difficult to distinguish by the discriminator. This mutual game process gradually improves the authenticity and detail expression ability of the generated images. However, relying solely on the basic generator loss may cause the generator to be unbalanced in learning different conditional features. Therefore, the present invention introduces a gradient normalization regularization term to balance the learning speed and quality of the generator on different conditional features. The purpose of this regularization term is to prevent the generator from learning too fast or too slow on certain feature layers by adjusting the gradient contribution of different feature layers. In the formula This balance strategy is expressed by calculating the gradient norm ratio of the generator at different feature layers, so that the generator can automatically adjust the importance weight α of each feature layer. k , thereby ensuring that all feature layers can be properly learned during training. Here, ∈ is a small constant used to prevent division by zero errors while ensuring the stability of gradient calculations. Through this gradient normalization method, the generator can achieve balanced learning at different feature scales, making the model's response to various surface change patterns more consistent and stable. In addition, in order to ensure that the generated image is highly consistent with the real image in the feature space, the generator's loss function also includes a reconstruction loss term. This reconstruction loss term uses the L2 norm to measure the difference between the generated image and the real image in the multi-scale conditional feature space. The formula described It is used to optimize this goal. By minimizing the Euclidean distance between the generated image and the real image in the feature space, this loss term ensures that the generator can not only generate realistic visual effects, but also approximate the distribution of real data in a higher-dimensional feature space. In other words, this reconstruction loss effectively forces the generator to consider not only the overall visual similarity when generating images, but also ensure that the detailed features are consistent with the real image. This is especially important for land change detection systems, because subtle changes in the surface are often the key to identification and early warning of the detection system.

[0082] Example 9: Loss function of the discriminator As shown below:.

[0083]

[0084] Among them, p data (I real ) is the probability distribution of the real image; 1-D(G(z, P′, A), P′, A) represents the authenticity judgment output of the discriminator on the generated image, and the probability that the generated image is false is obtained;

[0085] is the first discriminator loss, which is used to maximize the probability that the discriminator correctly identifies the real image; is the second discriminator loss, which is used to maximize the probability that the discriminator correctly identifies the generated image; is a multi-scale contrast loss term that enhances the sensitivity of the discriminator to subtle changes by using the difference in the discriminator outputs at different scales; p and q are both the norms of the contrast amount.

[0086] Specifically, the core part of the generator loss function is the basic generator loss, which implements the game between the generator and the discriminator through the adversarial learning mechanism. Specifically, the goal of the generator is to generate images that are as close to the real image as possible, making it difficult for the discriminator to distinguish between the real and the fake. To this end, the generator tries to minimize the discriminator's misjudgment rate of the generated images, that is, by minimizing the expression This loss term represents the adversarial goal of the generator, which is to maximize the probability that the generated image is recognized as a real image by the discriminator. In this way, the generator continuously adjusts its parameters during the training process, trying to "cheat" the discriminator to generate more realistic images. Under the framework of generative adversarial networks, this adversarial learning mechanism prompts the generator and the discriminator to compete and evolve together during the training process, and ultimately enables the generator to simulate high-precision surface changes. However, adversarial loss alone is not enough to guarantee the comprehensive learning effect of the generator. In the application of generative adversarial networks, different feature layers may contribute differently to the learning of the generator, so a mechanism is needed to balance the learning speed and effect of the generator on different feature layers. To this end, the present invention introduces a gradient normalization regularization term to adjust the learning strength of the generator on different conditional features. The core idea of ​​this regularization term is to balance the learning process of the generator by measuring the gradient contribution of different feature layers. The gradient normalization part in the formula It aims to control the learning rate of each feature layer to prevent some feature layers from over-learning and other feature layers from under-learning. In this way, the generator can learn evenly in the multi-scale feature space, so as to better capture the overall picture of surface changes. Here ∈ is a constant to prevent division by zero, ensuring the stability and numerical security of the calculation. The introduction of this regularization term effectively avoids the problem of gradient explosion or disappearance, making the learning process of the generator more stable and efficient. In addition, in order to further enhance the consistency between the generated image and the real image in the feature space, the generator loss function also includes a reconstruction loss term. This reconstruction loss term measures the difference between the generated image and the real image in the multi-scale conditional feature space through the L2 norm. Specifically, the reconstruction loss term It aims to minimize the Euclidean distance between the generated image and the real image in the feature space. This distance measurement method ensures that the generator not only generates realistic images in terms of visual effects, but also maintains similarity with the real image in the high-dimensional feature space. By optimizing this reconstruction loss, the generator is forced to not only imitate the overall visual features when generating images, but also capture the subtle surface change features. This is crucial for land change detection systems, because surface changes often show different characteristics at different scales, and the slight changes in these features may be exactly what the detection system needs to focus on identifying and analyzing. Through the joint action of the above three parts, the generator's loss function can fully guide the generator's learning direction and strategy during the training process. The basic generator loss ensures that the generator always seeks the optimal solution in the confrontation with the discriminator, thereby generating realistic images; the gradient normalization regularization term balances the learning rates of different feature layers to avoid overfitting or underfitting problems; the reconstruction loss term further ensures that the generated image maintains a high degree of consistency with the real image in the feature space. The design of this multiple loss strategy enables the generator to effectively learn and simulate various surface changes when processing complex remote sensing data while maintaining an efficient learning process. Ultimately, the design innovation of this loss function is that it can solve the challenges faced by traditional generative adversarial networks when processing high-dimensional and multi-scale data, and improve the overall performance and accuracy of the land change detection system of the present invention.

[0087] Embodiment 10: In the land change detection part, the change detection is performed based on the generated image pairs to obtain the change detection score by the following formula:

[0088]

[0089] Where, is the final change detection score, which is used to quantify the changes between the two generated images. The higher the score, the more significant the change. N represents the total number of pixels. G1(x i ,y i ) and G2(x i ,y i ): respectively represent different times at the position (x i ,y i ) the pixel values ​​of the first and second generated images; and Respectively represent the first and second generated images at position (x i ,y i ); γ: a weight parameter with a value range of [0, 1], which is used to balance the importance of gradient change and Euclidean distance in change detection.

[0090] Specifically, the two main comparison indicators are image gradient difference and pixel value difference. These two indicators represent different dimensions of image detail change and overall change respectively. It is mainly used to capture changes in edges and details in images. Gradient is a measure of image grayscale changes, which can sensitively reflect changes in edges and textures in images. Therefore, by comparing the gradient differences between the two generated images, the system can better detect subtle but significant edge information in surface changes, such as changes in building outlines, road widening or contraction, etc. This sensitivity to edge changes is particularly important for capturing high-precision land change information, because edge changes are often the most intuitive manifestation of surface changes. On the other hand, the pixel value difference || G1(x i ,y i )-G2(x i ,y i )||2 is used to evaluate the change in the overall brightness or color of the image, which may reflect a larger range of surface changes, such as a decrease or increase in vegetation cover, large-scale land development or flooding. By calculating the difference in Euclidean distance at each pixel position of the generated image, the system can quantify the magnitude of the change in the overall visual features between the two images. The introduction of pixel value differences enables the detection process to capture large-scale macroscopic change features, ensuring the effectiveness of the system in large-scale change detection. The weight parameter γ in the formula is used to balance between gradient changes and pixel value changes. The parameter γ ranges between 0 and 1, which determines the relative importance of the two differences in the final change detection score. By adjusting γ, the system can find the best balance between the sensitivity and robustness of detection according to actual application requirements. For example, in some tasks, if the characteristics of surface changes are mainly reflected in the changes in edges and details, a higher γ value can be selected to increase the weight of gradient changes; in another case, if the focus is on the changes in overall surface features, the γ value can be reduced to increase the proportion of pixel value differences in change detection. This flexibility in weight adjustment enables the change detection method of the present invention to adapt to different types of remote sensing data and detection tasks, thereby providing a highly customizable detection solution. In addition, the normalization factor in the formula It is used to average the difference in changes at all pixel positions to obtain a global change detection score. This score provides a unified quantitative indicator for the change at each pixel position, allowing the system to make an overall assessment of the degree of change in the entire image. In this way, the CDS score can not only reflect the details of local changes, but also summarize the overall changes in the surface, providing a powerful decision support tool for land management and environmental monitoring.

[0091] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A land change detection system based on generative adversarial networks, characterized in that: The system includes: a multi-source data fusion and feature extraction part, which is used to perform high-dimensional fusion of multi-source remote sensing data, consider time series and spatial distribution, ensure the spatiotemporal consistency and integrity of information, and obtain fused data; an adaptive convolution kernel design is used to capture non-stationary spatiotemporal features in the fused data; based on the non-stationary spatiotemporal features, a multi-scale feature pyramid construction method is used to ensure accurate detection of changes at different scales, and pyramid features at different scales are obtained through downsampling and upsampling operations of multiple layers of features; a spatiotemporal attention mechanism is introduced to adaptively adjust the importance of pyramid features to obtain attention weights; a generative adversarial network part is used to design a conditional generative adversarial network, and conditional features are obtained based on pyramid features and attention weights. The generator generates images after simulated changes, and the discriminator guides the improvement of the generator by evaluating the authenticity of these images; a land change detection part is used to consider the gradient change and Euclidean distance of the generated image, and change detection is performed based on the generated image pairs to obtain a change detection score.

2. The land change detection system based on generative adversarial network according to claim 1, characterized in that: In the multi-source data fusion and feature extraction part, multi-source remote sensing data is fused in high dimensions, taking into account time series and spatial distribution to ensure the temporal and spatial consistency and integrity of information. The process of obtaining fused data specifically includes: collecting raw data from n different remote sensing data sources; using interpolation and resampling to align all raw data in time and space to ensure that all raw data are on the same spatiotemporal grid, with equal weight for each data source, and the obtained fused data is expressed using the following formula: in, is the fused data set, which is a function of time t and spatial position (x, y); x is the X-axis coordinate; y is the Y-axis coordinate; T is the total time span; Ω is the spatial domain; is the signal strength of the ith data source, τ is the time offset, u is the X-axis offset, v is the Y-axis offset, λ i is the time decay coefficient of the i-th data source; σ i is the spatial distribution standard deviation of the ith data source; αt i is the temporal resolution of the ith data source; P i is the periodic change period of the i-th data source; α i is the periodic variation intensity coefficient of the i-th data source; is a Gaussian kernel used to simulate spatial correlation, σ i is the correlation range parameter, which controls the range of the control correlation; Used to handle differences in temporal resolution between different data sources.

3. The land change detection system based on generative adversarial network according to claim 2, characterized in that: In the multi-source data fusion and feature extraction part, adaptive convolution kernel design is used to capture the non-stationary spatiotemporal features F in the fused data through the following formula: Among them, Z is the normalization constant, which ensures that the integral of the convolution kernel is 1; σ t is the preset time dimension adaptive standard deviation; σ s Adaptive standard deviation for preset spatial dimensions.

4. The land change detection system based on generative adversarial network as claimed in claim 3, characterized in that: In the multi-source data fusion and feature extraction part, pyramid features of different scales are obtained through the following formula: Among them, P l is the pyramid feature corresponding to the lth level scale; DownSample l represents the lth downsampling operation; is the Laplace operator; represents the kth upsampling operation; P k is the pyramid feature corresponding to the k-th level scale.

5. The land change detection system based on generative adversarial network as described in claim 4 is characterized in that: In the multi-source data fusion and feature extraction part, the spatiotemporal attention mechanism is introduced to adaptively adjust the importance of pyramid features. The specific process of obtaining the attention weight includes: calculating the spatiotemporal correlation matrix according to the features of each layer of the pyramid through the following formula: Among them, C ij is the element in the i-th row and j-th column of the spatiotemporal correlation matrix; t i , t j are the time of feature i and feature j respectively; the attention weight of each layer of pyramid features is calculated using the spatiotemporal correlation matrix: Among them, A l is the attention weight of the l-th level scale pyramid feature at time t and spatial position (x, y); F i , F j is the feature of feature i and feature j at time t and spatial position (x, y); σ F is the importance parameter for controlling the similarity of eigenvalues ​​and is the set value.

6. The land change detection system based on generative adversarial network as described in claim 5 is characterized in that: The generator formula of the conditional generative adversarial network designed in the generative adversarial network part is expressed as follows: Among them, z is a random noise vector, which follows a normal distribution. P′ l is the conditional feature of the lth layer, which is calculated by the following formula: P′ l =A l ·P l ; A l is the attention weight of the l-th level scale pyramid feature; G(z, P′, A) is the simulated changed image generated by the generator; W G and b G are the weight matrix and bias vector of the generator respectively; φ(·) is the nonlinear activation function, which is the ReLU function or the LeakyReLU function; σ(·) is the output layer activation function, which is the tanh function or the sigmoid function; U is the weight matrix of the conditional input of the generator, which is used to fuse the conditional features and noise; A is the attention weight matrix; z is the noise; the output of the generator is a simulated image, which is as similar as possible to the real image in high-dimensional space.

7. The land change detection system based on generative adversarial network as described in claim 6 is characterized in that: The formula of the discriminator is as follows: Among them, I is the input image, which is the real image I real Or the simulated image G(z, P′, A) generated by the generator; D(I, P′, A) is the probability output by the discriminator, indicating the authenticity of the input image I; W D and b D are the weight matrix and bias vector of the discriminator respectively; V is the weight matrix of the conditional input of the discriminator; D(I, P′, A) is the probability of the discriminator output, indicating the authenticity of the input image I.

8. The land change detection system based on generative adversarial network as described in claim 7 is characterized in that: The loss function of the generator As shown below: Among them, p z (z) is the probability distribution of random noise z; is the expected operator; P′ is the conditional feature set; is the basic generator loss, which is used to maximize the probability that the generated image is recognized as a real image by the discriminator; is a gradient normalization regularization term used to balance the learning speed of the generator on different conditional features; and are the gradients of the generator and conditional features, α k is the adaptive weight of each feature layer, ∈ is a constant to prevent division by zero; It is a reconstruction loss term, which ensures the consistency between the generated image and the real image in the conditional feature space. The L2 norm is used to measure the difference between the generated image and the real image; ∥∥2 represents the L2 norm operation; represents the loss part corresponding to the conditional features of the kth layer; ∥∥ represents the operation of finding the determinant of the matrix or the modulus of the vector; p P′ (P′) is the probability distribution of conditional feature P′; I real (P′ l ) is the conditional feature of the real image at the lth level scale.

9. The land change detection system based on generative adversarial network as claimed in claim 8, characterized in that: Loss function of the discriminator As shown below:. Among them, p data (I real ) is the probability distribution of the real image; 1-D(G(z, P′, A), P′, A) represents the authenticity judgment output of the discriminator on the generated image, and the probability that the generated image is false is obtained; is the first discriminator loss, which is used to maximize the probability that the discriminator correctly identifies the real image; is the second discriminator loss, which is used to maximize the probability that the discriminator correctly identifies the generated image; is a multi-scale contrast loss term that enhances the sensitivity of the discriminator to subtle changes by using the difference in the discriminator outputs at different scales; p and q are both the norms of the contrast amount.

10. The land change detection system based on generative adversarial network according to claim 9, characterized in that: In the land change detection part, the change detection is performed based on the generated image pairs and the change detection score is obtained by the following formula: Where, is the final change detection score, which is used to quantify the changes between the two generated images. The higher the score, the more significant the change. N represents the total number of pixels. G1(x i ,y i ) and G2(x i ,y i ): respectively represent different times at the position (x i ,y i ) the pixel values ​​of the first and second generated images; and Respectively represent the first and second generated images at position (x i ,y i ); γ: a weight parameter with a value range of [0, 1], which is used to balance the importance of gradient change and Euclidean distance in change detection.

Images