Mountain area cultivated land non-grain remote sensing intelligent monitoring method for zero sample change detection

Abstract

The invention provides a mountain cultivated land non-grain remote sensing intelligent monitoring method based on zero sample change detection, and the method comprises the steps: collecting a high-resolution optical image, a cultivated land vector and economic crop data, constructing a cultivated land non-grain change detection typical sample library containing seven change types, calculating the change feature vector of each change type, and calculating the change feature vector of each change type; semantic segmentation is carried out on front and later images by using an SAM model, a semantic information change detection operator for spatial regularization diffusion learning clustering post-processing and dual-time-phase potential space matching is designed, and non-grain pattern spot extraction of the mountain cultivated land based on a zero sample strategy is realized. According to the method, only a small number of typical samples are needed to construct the sample library, large-scale training sample marking is avoided, the cost is low, the efficiency is high, the segmentation precision of the broken farmland in the mountainous area is improved through the space regularization diffusion learning clustering post-processing algorithm, the farmland non-grain detection precision is improved through the semantic information change detection operator matched with the double-time-phase potential space, and the method is suitable for popularization and application. And a new solution is provided for fine identification of cultivated land and grain safety monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning change detection, and particularly relates to a zero-sample change detection remote sensing intelligent monitoring method for mountainous cultivated land non-food production. BACKGROUND

[0002] Cultivated land non-food production monitoring mainly refers to that in the grain planting area, forest fruits, forage grasses, fish, earth, and other facilities cannot be planted, and other facilities cannot be built. Such non-food production patches need to be extracted. Traditional cultivated land non-food production monitoring mainly relies on manual field verification and cultivated land area and type reporting by villages. This method has two main drawbacks: first, manual field verification cannot be fully carried out and can only be sampled and verified, which is time-consuming and labor-intensive, and the verification accuracy is directly related to the operation proficiency and responsibility of the operator; second, the reporting by villages has the problem of false reporting, and farmers have the mentality of taking chances and concealing information about planting economic crops or breeding.

[0003] Traditional cultivated land non-food production monitoring relying on manual field verification obviously cannot meet the current demand for large-scale, high-frequency, and fine-grained cultivated land monitoring. In recent years, with the improvement of satellite remote sensing in spatial resolution and spectral resolution, and the vigorous development of remote sensing big data intelligent analysis technology, new opportunities have been brought to cultivated land non-food production monitoring. There are mainly two ways to use remote sensing intelligent technology for cultivated land non-food production monitoring: one is to make cultivated crop remote sensing intelligent interpretation samples, train and build a cultivated crop remote sensing fine-grained intelligent recognition model, intelligently interpret each period of multi-temporal images, and then compare and analyze to extract cultivated land non-food production patches; the other is to make cultivated land non-food production change detection samples, train and build a change detection model, and extract cultivated land non-food production patches from two images. Both of these two deep learning driven ideas need to label large-scale and high-quality training samples. Due to the fragmented and scattered characteristics of mountainous cultivated land, and the complexity of surface coverage types in mountainous areas, cultivated land and garden grassland are mixed, and intercropping, relay cropping, and interplanting are very common in cultivated land. It is very difficult to label large-scale and high-quality cultivated crop or non-food production change detection sample data at the pixel level in actual business work, which requires a lot of manpower and a long time. In recent years, in view of the practical difficulties of large-scale deep learning training sample making, zero-sample learning has been proposed by scholars to identify and segment target objects without training samples. SUMMARY

[0004] Therefore, it is necessary to provide a zero-sample change detection remote sensing intelligent monitoring method for mountainous cultivated land non-food production in view of the above technical problems.

[0005] A zero-sample change detection remote sensing intelligent monitoring method for mountainous cultivated land non-food production, comprising the following steps: Step S1, obtaining high-resolution optical images of different years of the region to be tested, cultivated land survey monitoring vectors of the corresponding years, and economic crop survey data of the next year; performing geometric precise correction and pixel-level position registration on the high-resolution optical images to obtain image data of the previous period and the later period; Step S2, obtaining change characteristics of each change type in the cultivated land non-grain change detection typical sample library; wherein the change types include: cultivated land changing into economic crops, cultivated land changing into forest land, cultivated land changing into nursery, cultivated land changing into grassland, cultivated land changing into pit pond, cultivated land changing into building, and cultivated land changing into excavation land; Step S3, performing sliding cutting on the image data of the previous period and the later period to obtain an image small piece group; wherein the image small piece group includes one previous period image small piece and one later period image small piece; performing segmentation on the image small piece group according to a SAM model and performing small piece splicing after segmentation to obtain a segmentation result of the image; Step S4, performing clustering post-processing on the segmentation result of the image by a spatial regularization diffusion learning clustering algorithm to obtain an optimized segmentation clustering result; Step S5, performing change map extraction on the optimized segmentation clustering result by a semantic information change detection operator of double-time-phase latent space matching to obtain a change map and calculate change characteristics; Step S6, calculating a Pearson correlation coefficient according to the extracted change map and change characteristics and the change characteristics of each change type in the typical sample library to obtain a map change type.

[0006] In one of the embodiments, before step S2, it further includes: Filtering cultivated land changing into non-cultivated land map from the cultivated land survey monitoring vectors and the economic crop survey data by using overlay analysis, obtaining typical map of cultivated land changing into non-cultivated land, and making a cultivated land non-grain change detection typical sample library; Obtaining a cultivated land non-grain change detection typical sample library and calculating change characteristics of each change type in the cultivated land non-grain change detection typical sample library.

[0007] In one of the embodiments, obtaining a cultivated land non-grain change detection typical sample library and calculating change characteristics of each change type in the cultivated land non-grain change detection typical sample library includes: Calculating change characteristics of a typical sample map under a change type, wherein the change characteristics of the typical sample map include: pixel-level change characteristics and object-level change characteristics: ; ; ; ; ; ; ; ; ; ; in, Indicates the first A typical example of pixel-level variation characteristics of image patches The first The difference between the mean values ​​of the histograms of the red, green, and blue bands in the preceding and following image data within a typical sample patch. The first The difference in standard deviations of the red, green, and blue band histograms of pre- and post-image data within a typical sample patch. Indicates the first Typical example image patch object-level variation characteristics The first The differences in roughness, contrast, directionality, linearity, regularity, and coarseness texture features in the red, green, and blue bands of the image data from different periods within a typical sample patch. The first The differences between the morphological top cap transformation features and the morphological bottom cap transformation features of the red, green, and blue bands in the image data of the previous and later stages of a typical sample patch. These represent the differences in roughness for the red, green, and blue bands, respectively. These represent the differences in contrast between the red, green, and blue bands, respectively. These represent the differences in directionality among the red, green, and blue bands, respectively. These represent the differences in linearity among the red, green, and blue bands, respectively. These represent the regularity differences between the red, green, and blue bands, respectively. These represent the differences in coarseness for the red, green, and blue bands, respectively. These represent the differences in morphological top-hat transformation characteristics for the red, green, and blue bands, respectively. These represent the differences in morphological cap transformation characteristics for the red, green, and blue bands, respectively. The variation characteristics of the typical sample patch yield the variation characteristics of the variation type: ; ; ; in, This represents the average value of pixel-level variation features. This represents the number of all typical sample patches under the CT-type variation. Indicates the first A typical example image patch, Indicates object-level change characteristics. This represents the change characteristics of the CT-th type of change.

[0008] In one embodiment, step S4 includes: The kernel density is calculated based on the segmentation results of the image, and representative pixels are selected according to the preset kernel density requirements. A spatially regularized KNN nearest neighbor graph is constructed based on the representative pixels to obtain a reduced spatially regularized diffusion graph. Based on the reduced spatial regularization diffusion map, K pixels are located as cluster pattern centers. Cluster labels are propagated to unlabeled pixels in the reduced spatial regularization diffusion map. The cluster labels of the remaining pixels are determined based on the cluster labels of the pixels in the reduced spatial regularization diffusion map, thus obtaining the optimized segmentation and clustering results.

[0009] In one embodiment, the kernel density is calculated based on the segmentation results of the image, and representative pixels are selected according to a preset kernel density requirement. A spatially regularized KNN nearest neighbor graph is constructed based on the representative pixels to obtain a reduced spatially regularized diffusion graph, including: Calculate the kernel density using the following formula: ; in, Represents the pixels within each patch object nuclear density, express The Middle 1 pixel, Represents pixels The set of nearest neighbor pixels within a radius of R pixels in the segmentation result, determined by Euclidean distance. , Representing pixels Pixel values ​​in the red band of the image, , Representing pixels Pixel values ​​in the green band of the image, , Representing pixels Pixel values ​​in the blue band of the image, This represents a scaling factor that controls the radius of interaction between pixels. Represents the regularization factor. make sure ; Within each patch object, the kernel density of each pixel is sorted in descending order, and the top k pixels are selected as the representative pixels of the patch object. The spatially regularized KNN nearest neighbor graph is constructed using the following formula, resulting in a reduced spatially regularized diffusion graph: ; in, This indicates a reduced spatial regularization diffusion map. Indicates the total number of objects in the map. Indicates the first A single image object, The number of pixels in .

[0010] In one embodiment, K pixels are located as cluster pattern centers based on the reduced spatial regularization diffusion map, cluster labels are propagated to unlabeled pixels in the reduced spatial regularization diffusion map, and cluster labels of the remaining pixels are determined based on the cluster labels of the pixels in the reduced spatial regularization diffusion map, resulting in optimized segmentation and clustering results including: The cluster pattern center representation value is calculated according to the following formula, and the K pixels with the largest cluster pattern center representation values ​​are selected as the cluster pattern centers: ; ; in, Represents the cluster pattern center representation value. Indicates time Time Pixel Compared with pixels in the reduced spatial regularization diffusion map With higher density The diffusion distance between nearest neighbors The time calculation is based on the spatially regularized diffusion map and Markov diffusion process. Time Pixel and Medium pixel diffusion distance, This represents a reduced spatial regularization diffusion plot; Cluster labels are propagated to unlabeled pixels in the reduced spatial regularization diffusion map according to the following formula: ; ; in, The cluster label function for pixel X. This indicates reducing the number of unlabeled pixels in the spatially regularized diffusion map. Indicates distance The most recent cluster center pixel, This represents the parameter that minimizes the function value. This indicates a reduced spatial regularization diffusion map. representing a cluster mode center, the value of K is 1, 2, 3, …, K, representing a pixel and at time diffusion distance; The correlation degree value is calculated according to the following formula, and the cluster label of the pixel with the maximum correlation degree value is selected: ; wherein, representing the correlation degree value, representing the unmarked cluster label pixel and the Mahalanobis distance of each marked cluster label pixel in the reduced spatial regularization diffusion map, representing the pixel its own kernel density estimation value; Merge the adjacent and cluster label same graph patches to obtain the optimized segmentation clustering result.

[0011] In one of the embodiments, step S5 comprises: Extracting the latent features of the pre-image and the post-image through the image encoder of the SAM model, and averaging the latent features in each channel in each graph patch of the optimized segmentation clustering result to obtain a latent feature vector; Calculating a graph patch change confidence score according to the latent feature vector to obtain the change degree of the pre-image and post-image graph patch semantics; Obtaining a graph patch change confidence score threshold, and in response to the graph patch change confidence score being greater than the graph patch change confidence score threshold, taking the graph patch as the extracted change graph patch.

[0012] In one of the embodiments, calculating a graph patch change confidence score according to the latent feature vector to obtain the change degree of the pre-image and post-image graph patch semantics comprises: Calculating the graph patch change confidence score according to the following formula: ; ; wherein, representing the graph patch change confidence score, representing the graph patch change confidence score of any graph patch of the pre-image small piece, representing the graph patch change confidence score of any graph patch of the post-image small piece, representing the graph patch change confidence score of any graph patch, representing the graph patch serial number, respectively representing the latent feature vectors of the corresponding graph patches at the same position of the pre-image and the post-image.

[0013] In one of the embodiments, the step S6 comprises: The Pearson correlation coefficient is calculated according to the following formula: ; wherein, represents the Pearson correlation coefficient, represents the covariance, represents the eigenvector of the indexth change patch, represents the variance of the change feature of the CTth change type; A correlation coefficient threshold is obtained, and in response to the Pearson correlation coefficient being greater than the correlation coefficient threshold, the type of the change patch is defined as the change type corresponding to the change feature.

[0014] A zero-sample change detection mountainous cultivated land non-food remote sensing intelligent monitoring system is used to implement the zero-sample change detection mountainous cultivated land non-food remote sensing intelligent monitoring method as described above, and comprises: A data acquisition module is configured to acquire high-resolution optical images of different years in a to-be-detected region, cultivated land survey and monitoring vectors of corresponding years, and economic crop survey data of the next year; geometric precise correction and pixel-level position registration are performed on the high-resolution optical images to obtain image data of the early stage and the later stage. A change sample acquisition module is configured to acquire change features of each change type in a typical sample library for cultivated land non-food change detection; wherein the change types include: cultivated land changing into economic crops, cultivated land changing into forest land, cultivated land changing into nursery, cultivated land changing into grassland, cultivated land changing into pit pond, cultivated land changing into building, and cultivated land changing into excavation land. An image segmentation module is configured to perform sliding cropping on the image data of the early stage and the later stage to obtain an image patch group; wherein the image patch group comprises one early-stage image patch and one later-stage image patch; the image patch group is segmented according to a SAM model and the segmented patches are spliced to obtain a segmentation result of the image. A clustering optimization module is configured to perform post-processing clustering on the segmentation result of the image through a spatial regularization diffusion learning clustering algorithm to obtain an optimized segmentation clustering result. A change extraction module is configured to perform change patch extraction on the optimized segmentation clustering result through a semantic information change detection operator of double-time-phase latent space matching to obtain change patches and calculate change features. A type acquisition module is configured to calculate a Pearson correlation coefficient according to the change patches and the change features and change features of each change type in the typical sample library to obtain patch change types.

[0015] Compared with the prior art, the advantages and beneficial effects of the present application are that the present application can ingeniously combine the open source SAM model with the innovative clustering algorithm, break through the bottleneck of traditional full supervised learning relying on large-scale labeled samples, and only need a small amount of typical samples to construct a change detection sample library, greatly reducing the cost of manpower and material resources; the spatial regularization diffusion learning clustering algorithm is used to optimize the segmentation result, effectively dealing with the problem of fragmentation of cultivated land in mountainous areas and complex land cover, and improving the accuracy of polygon segmentation; the semantic information change detection operator of double temporal phase potential space matching is designed, the polygon change confidence is quantified by using the cosine distance and the bidirectional matching strategy, and the temporal symmetry and robustness of change detection are enhanced; the overall process deeply integrates pixel-level and object-level feature analysis, realizes the intelligentization of the whole process from change polygon extraction to type identification, provides a technical scheme with low cost, high efficiency and high precision for monitoring of non-grainization of cultivated land in mountainous areas, and has important application value for fine protection of cultivated land and security of food safety. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 FIG. 1 is a flowchart of a mountainous cultivated land non-grainization remote sensing intelligent monitoring method of zero sample change detection in an embodiment; Figure 2A FIG. 2 is a schematic diagram of cultivated land changing into building and dumping in an embodiment; Figure 2B FIG. 3 is a schematic diagram of cultivated land changing into pit in an embodiment; Figure 3 FIG. 4 is a structural schematic diagram of a mountainous cultivated land non-grainization remote sensing intelligent monitoring system of zero sample change detection in an embodiment. DETAILED DESCRIPTION

[0017] Before the specific embodiment of the present application is described, the overall concept of the present application is described as follows: The present application is mainly developed based on the traditional cultivated land non-grainization monitoring process. The traditional cultivated land non-grainization monitoring relying on manual field verification obviously cannot meet the current cultivated land protection work requirements for wide range, high frequency and fine cultivated land monitoring.

[0018] In view of the deficiencies of the prior art, the present application provides a mountainous cultivated land non-grainization remote sensing intelligent monitoring method of zero sample change detection, to solve the problems of high cost and low efficiency of labeling large-scale cultivated land non-grainization change detection samples due to complex mountainous terrain, fragmented cultivated land, and intercropping and relay cropping in cultivated land, and the insufficient performance and robustness of traditional intelligent models in the prior art, to find out the change polygons by using the zero sample change detection strategy, and then to filter out the change polygons of cultivated land non-grainization by extracting the features such as color, texture and shape of the pre and post images in the change polygons, and determining the similarity with the cultivated land non-grainization change detection sample library.

[0019] Having introduced the overall concept of the present application, in order to make the purpose, technical scheme and advantages of the present application more clear and understandable, the present application will be further described in detail below through specific embodiments combined with the drawings.

[0020] It should be noted that, unless otherwise defined, technical or scientific terms used in one or more embodiments of the present application should be understood as their ordinary meaning to a person having ordinary skill in the art to which the present application belongs. The terms "first", "second" and similar terms used in one or more embodiments of the present application do not represent any order, number or importance, but are only used to distinguish different components. The terms "include" or "contain" and similar terms mean that the elements or objects before the terms cover the elements or objects listed after the terms and their equivalents, without excluding other elements or objects. The terms "connect" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "up", "down", "left", "right" and the like are only used to represent relative positional relationships, which can change accordingly when the absolute position of the described object changes.

[0021] In one embodiment, as shown in Figure 1 a mountainous farmland non-food remote sensing intelligent monitoring of zero sample change detection is provided, comprising the following steps: Step S1, obtaining high-resolution optical images of the to-be-measured region in different years, cultivated land survey and monitoring vectors corresponding to the years, and economic crop survey data of the following year; performing geometric precise correction and pixel-level position registration on the high-resolution optical images to obtain image data of the previous and later periods.

[0022] Specifically, high-resolution optical images of the to-be-measured region in two years, cultivated land survey and monitoring vectors corresponding to the two years, and economic crop survey data of the following year are collected and obtained, the two years before and after are taken as the previous and later periods, the images are geometrically corrected and pixel-level position registered to obtain image data of the previous and later periods, and cultivated land that becomes non-cultivated land is filtered out from the cultivated land survey and monitoring vectors of the two years and the economic crop survey data of the following year through overlay analysis.

[0023] In the present embodiment, 0.5-meter resolution images of Beibei District, Chongqing City are collected, the time phase of the previous image is July 2020, and the time phase of the later image is August 2024, the cultivated land survey and monitoring vectors of 2020 are derived from the 2020 land survey and monitoring results, the cultivated land survey and monitoring vectors of 2024 are derived from the 2024 land change survey results, and the economic crop survey data of 2024 is derived from the cultivated forest space special investigation results.

[0024] Step S2, obtaining the change characteristics of each change type in the cultivated land non-grain change detection typical sample library; wherein, the change types include: cultivated land changing into economic crops, cultivated land changing into forest land, cultivated land changing into nursery, cultivated land changing into grassland, cultivated land changing into pit pond, cultivated land changing into building, and cultivated land changing into excavated land.

[0025] Specifically, the change characteristics of each change type in the constructed cultivated land non-grain change detection typical sample library are obtained, and the change types of cultivated land non-grain change detection include 7 types in total, wherein the types of cultivated land changing into non-cultivated land include cultivated land changing into economic crops, cultivated land changing into forest land, cultivated land changing into nursery, cultivated land changing into grassland, cultivated land changing into pit pond, cultivated land changing into building, and cultivated land changing into excavated land, and 5-10 typical samples of each change type are selected.

[0026] On this basis, step S2 further includes: filtering out the cultivated land changing into non-cultivated land plot from the cultivated land survey and monitoring vector and economic crop survey data by using superposition analysis, obtaining the typical plot of cultivated land changing into non-cultivated land, and making the cultivated land non-grain change detection typical sample library; obtaining the cultivated land non-grain change detection typical sample library, and calculating the change characteristics of each change type in the cultivated land non-grain change detection typical sample library.

[0027] Specifically, the cultivated land non-grain change detection typical sample library includes the typical sample plot of cultivated land changing into non-cultivated land and the corresponding image data of the previous and later periods.

[0028] In the embodiment, based on the high-resolution optical images of the previous and later periods, the typical sample plot of cultivated land changing into non-cultivated land is selected by visual interpretation, the marking information is generated, the marking information is received, and 10 typical sample plots of cultivated land changing into economic crops are found out from the images of the previous and later periods in combination with the later economic crop survey data, so as to make the cultivated land non-grain change detection typical sample library.

[0029] Further, the images used for making the cultivated land non-grain change detection sample library in the embodiment are located in Longfengqiao Street, Xiemazhen, Shijiangling Town, and Caijiagang Town of Beibei District.

[0030] On this basis, obtaining the cultivated land non-grain change detection typical sample library and calculating the change characteristics of each change type in the cultivated land non-grain change detection typical sample library include: calculating the change characteristics of the typical sample plot under the change type, and the change characteristics of the typical sample plot include: pixel-level change characteristics and object-level change characteristics: ; ; ; ; ; ; ; ; ; ; in, Indicates the first A typical example of pixel-level variation characteristics of image patches The first The difference between the mean values ​​of the histograms of the red, green, and blue bands in the preceding and following image data within a typical sample patch. The first The difference in standard deviations of the red, green, and blue band histograms of pre- and post-image data within a typical sample patch. Indicates the first Typical example image patch object-level variation characteristics The first The differences in roughness, contrast, directionality, linearity, regularity, and coarseness texture features in the red, green, and blue bands of the image data from different periods within a typical sample patch. The first The differences between the morphological top cap transformation features and the morphological bottom cap transformation features of the red, green, and blue bands in the image data from different periods within a typical sample patch. These represent the differences in roughness for the red, green, and blue bands, respectively. These represent the differences in contrast between the red, green, and blue bands, respectively. These represent the differences in directionality among the red, green, and blue bands, respectively. These represent the differences in linearity among the red, green, and blue bands, respectively. These represent the regularity differences between the red, green, and blue bands, respectively. These represent the differences in coarseness for the red, green, and blue bands, respectively. These represent the differences in morphological top-hat transformation characteristics for the red, green, and blue bands, respectively. These represent the differences in morphological cap transformation characteristics for the red, green, and blue bands, respectively. The variation characteristics of the typical sample patch yield the variation characteristics of the variation type: ; ; ; in, This represents the average value of pixel-level variation features. indicates the number of all typical sample patches under the first CT type of change, indicates the first typical sample patch, indicates the object-level change feature, indicates the change feature of the first CT type of change.

[0031] Specifically, the change feature of each change type in the cultivated land non-grain change detection typical sample library is calculated, including the following processes: First, the change feature of the first typical sample patch under the first CT type of change is calculated, mainly including the pixel-level change feature and the object-level change feature, the pixel-level change feature mainly being the difference between the mean and the standard deviation of the histogram of the three bands of the pre-period and post-period images in the typical sample patch, and the object-level change feature mainly including the Tamura texture features and the morphological features of the three bands of the pre-period and post-period images in the typical sample patch, the Tamura texture features including roughness, contrast, directionality, linearity, regularity, and coarseness, and the morphological features including morphological top-hat transformation and morphological bottom-hat transformation; ; ; wherein, are the differences between the means of the red, green, and blue band histograms of the pre-period and post-period images in the typical sample patch, are the differences between the standard deviations of the red, green, and blue band histograms of the pre-period and post-period images in the typical sample patch, are the differences between the six Tamura texture features of the red, green, and blue bands of the pre-period and post-period images in the typical sample patch, including roughness, contrast, directionality, linearity, regularity, and coarseness, are the differences between the morphological top-hat transformation features and the morphological bottom-hat transformation features of the red, green, and blue bands of the pre-period and post-period images in the typical sample patch, and the calculation method is as follows: ; ; ; ; ; ; ; ; wherein, Roughness of the red, green and blue bands of the pre-stage image within the typical sample plot patch, Roughness of the red, green and blue bands of the post-stage image within the typical sample plot patch, the roughness includes but is not limited to the standard deviation of the gray level difference in the neighborhood of the pixel; Contrast of the red, green and blue bands of the pre-stage image within the typical sample plot patch, Contrast of the red, green and blue bands of the post-stage image within the typical sample plot patch, the contrast includes but is not limited to the second moment of the gray level histogram; Directionality of the red, green and blue bands of the pre-stage image within the typical sample plot patch, Directionality of the red, green and blue bands of the post-stage image within the typical sample plot patch, the directionality includes but is not limited to the peak detection of the gradient direction histogram; Linearity of the red, green and blue bands of the pre-stage image within the typical sample plot patch, Linearity of the red, green and blue bands of the post-stage image within the typical sample plot patch, the linearity includes but is not limited to the variance calculation based on the edge direction distribution; Regularity of the red, green and blue bands of the pre-stage image within the typical sample plot patch, Regularity of the red, green and blue bands of the post-stage image within the typical sample plot patch, the regularity includes but is not limited to the multi-scale autocorrelation analysis; Coarseness of the red, green and blue bands of the pre-stage image within the typical sample plot patch, Coarseness of the red, green and blue bands of the post-stage image within the typical sample plot patch, the coarseness includes but is not limited to the wavelet transform energy coefficient calculation; Morphological top-hat transform feature of the red, green and blue bands of the pre-stage image within the typical sample plot patch, Morphological top-hat transform feature of the red, green and blue bands of the post-stage image within the typical sample plot patch; Morphological bottom-hat transform feature of the red, green and blue bands of the pre-stage image within the typical sample plot patch, Morphological bottom-hat transform feature of the red, green and blue bands of the post-stage image within the typical sample plot patch.

[0032] Next, the change features of all the typical sample plot patches of the change type are calculated, and the average value is taken to obtain the change feature of the change type, and combined into a one-dimensional vector ; ; ; ; wherein Nct is the number of all the typical sample plot patches of the CTth change type.

[0033] Finally, the process of the above steps is calculated for each of the seven change types (a total of seven) to obtain the change feature vector of all seven change types 、 、 、 、 、 、 .

[0034] Step S3, sliding cropping is performed on the pre-stage and post-stage image data to obtain an image patch group; wherein the image patch group includes a pre-stage image patch and a post-stage image patch; the image patch group is segmented according to a SAM model and the segmented patches are spliced to obtain a segmentation result of the image.

[0035] Specifically, the pre-stage and post-stage images are slidingly cropped to produce patches, and a pre-selected sliding window (1024*1024 pixel size is selected in this embodiment) is used to crop the pre-stage and post-stage images in a left-to-right and top-to-bottom movement manner, and the movement step of the sliding window is 1024 pixels, that is, the image patches do not overlap, and finally Sm groups of image patches are obtained, each group of image patches including a pre-stage image patch and a post-stage image patch.

[0036] For each group of image patches, the pre-stage and post-stage image patches are segmented using an open-source SAM (Segment Anything Model) model to obtain pre-stage image patch segmentation results and post-stage image patch segmentation results, and the segmentation results of all pre-stage image patches are spliced to obtain a pre-stage image segmentation result , the segmentation results of all post-stage image patches are spliced to obtain a post-stage image segmentation result , and the segmentation result of the image includes the pre-stage image segmentation result and the post-stage image segmentation result.

[0037] Further, the images used for cropping and producing patches in this embodiment are located in Tianfu Town, Jingguan Town, Shuitu Town, and Fuxing Town of Beibei District, which are different from the spatial positions of the images used for producing the non-grain change detection sample library.

[0038] Step S4, the segmentation result of the image is clustered and post-processed by a spatial regularization diffusion learning clustering algorithm to obtain an optimized segmentation clustering result.

[0039] Specifically, the segmentation results of the two-stage images are clustered and post-processed by a spatial regularization diffusion learning clustering algorithm to obtain a pre-stage optimized image segmentation clustering optimization result and a post-stage optimized image segmentation clustering optimization result .

[0040] Based on this, the segmentation results of the image are clustered using a spatial regularized diffusion learning clustering algorithm to obtain optimized segmentation and clustering results, including: The kernel density is calculated based on the segmentation results of the image, and representative pixels are selected according to the preset kernel density requirements. A spatially regularized KNN nearest neighbor graph is constructed based on the representative pixels to obtain a reduced spatially regularized diffusion graph. Based on the reduced spatial regularization diffusion map, K pixels are located as cluster pattern centers. Cluster labels are propagated to unlabeled pixels in the reduced spatial regularization diffusion map. The cluster labels of the remaining pixels are determined based on the cluster labels of the pixels in the reduced spatial regularization diffusion map, thus obtaining the optimized segmentation and clustering results.

[0041] The kernel density is calculated based on the segmentation results of the image, and representative pixels are selected according to the preset kernel density requirements. A spatially regularized KNN nearest neighbor graph is constructed based on the representative pixels to obtain a reduced spatially regularized diffusion graph, including: Calculate the kernel density using the following formula: ; in, Represents the pixels within each patch object nuclear density, express A certain pixel in, Represents pixels The set of nearest neighbor pixels within a radius of R pixels in the segmentation result, determined by Euclidean distance. , Representing pixels Pixel values ​​in the red band of the image, , Representing pixels Pixel values ​​in the green band of the image, , Representing pixels Pixel values ​​in the blue band of the image, This represents a scaling factor that controls the radius of interaction between pixels. Represents the regularization factor. make sure ; Within each patch object, the kernel density of each pixel is sorted in descending order, and the top k pixels are selected as the representative pixels of the patch object. The spatially regularized KNN nearest neighbor graph is constructed using the following formula, resulting in a reduced spatially regularized diffusion graph: ; in, This indicates a reduced spatial regularization diffusion map. Indicates the total number of objects in the map. representing the first graphical region object, the number of pixels in .

[0042] According to the reduced spatial regularization diffusion map positioning K pixels as cluster mode centers, propagating cluster labels to unmarked pixels in the reduced spatial regularization diffusion map, determining cluster labels of the remaining pixels according to the cluster labels of the pixels in the reduced spatial regularization diffusion map, and obtaining an optimized segmentation clustering result, comprising: The cluster mode center representation value is calculated according to the following formula, and the K pixels with the largest cluster mode center representation value are selected as the cluster mode centers: ; ; Among them, represents the cluster mode center representation value, represents the diffusion distance between pixel and the pixel with higher density in the reduced spatial regularization diffusion map at time , is the diffusion distance between pixel and pixel in the reduced spatial regularization diffusion map at time , represents the reduced spatial regularization diffusion map; The cluster label is propagated to the unmarked pixels in the reduced spatial regularization diffusion map according to the following formula: ; ; Among them, represents the cluster label function of pixel x, represents the unmarked pixel in the reduced spatial regularization diffusion map, represents the nearest cluster center pixel, represents the parameter that minimizes the function value, represents the reduced spatial regularization diffusion map, represents the cluster mode center, the value of is 1, 2, 3,..., K, represents the diffusion distance between pixel and at time ; The correlation score is calculated using the following formula, and the cluster label of the pixel with the highest correlation score is selected: ; in, Indicates the degree of correlation. Indicates unlabeled clustered pixels With each labeled cluster tag pixel in the reduced spatial regularization diffusion map Mahalanobis distance, Represents pixels Its own kernel density estimate; Adjacent patches with the same cluster label are merged to obtain optimized segmentation and clustering results.

[0043] Specifically, the post-processing of the segmentation results of a certain period of image using the spatial regularized diffusion learning clustering algorithm mainly includes the following steps: Step 4.1.1, constructing a spatially regularized diffusion graph, including calculating kernel density estimation, selecting representative pixels, and constructing a spatially regularized KNN nearest neighbor graph, including the following steps.

[0044] Step 4.1.10: Treat each patch in the segmentation result image of the open-source SAM model as a patch object, with a total number of patch objects. , For the first Each patch object is used to calculate pixels within each patch object. Density(x) of the kernel: ; in This refers to a specific pixel within the image of the patch object. , For all pixels of the image, Indicates that pixel x is in The set of nearest neighbor pixels within a radius of R pixels in the middle (where the nearest neighbor radius is R pixels), y is... A certain pixel in, Z is a scaling factor that controls the interaction radius between pixels, and Z is a regularization factor that ensures... =1, , For each pixel Pixel values ​​in the red band of the image, , For each pixel Pixel values ​​in the green band of the image, , For each pixel Pixel values ​​in the blue band of the image.

[0045] Further, the total number of spot objects in the pre-stage image of the embodiment = 1364, and the total number of spot objects in the post-stage image of the embodiment = 1528, the nearest neighbor radius R of the embodiment is set to 10, and the scaling factor of the interaction radius between pixels is set to 6.

[0046] Step 4.1.11, in each spot object, the kernel density Density(x) of each pixel is arranged in descending order according to the kernel density requirement (select a larger kernel density), and the first k pixels are selected as the representative pixels of the spot object.

[0047] Step 4.1.12, using the selected representative pixels as the most representative high-density pixels, representative pixels are recorded as , and a KNN neighbor graph is established using the pixels in to realize the construction of a diffusion graph with reduced spatial regularization: ; Further, the embodiment selects 6820 representative pixels in the pre-stage image and 7640 representative pixels in the post-stage image.

[0048] Step 4.1.2, diffusion learning clustering is performed using the constructed diffusion graph with reduced spatial regularization, including positioning K pixels as clustering mode centers, propagating cluster labels to unmarked pixels in , determining the cluster labels of the remaining pixels according to the cluster labels of the pixels in , including the following steps.

[0049] Step 4.1.20, positioning K pixels in as clustering mode centers, i.e., representative pixel samples of potential clustering structure, and assigning unique cluster labels (label values are 1, 2, 3,..., K) to the K pixels, recording a certain one of the K pixels as , the corresponding label is lb , i.e. , the clustering mode center is the K pixel that maximizes , the clustering mode center of the K pixels is the highest density pixel in that is farthest from other high-density pixels in the diffusion distance; ; wherein is the diffusion distance between pixel x and the pixel v in with higher density between the nearest neighbors at time t, ​The pixel x at time t is calculated based on the diffusion map with reduced spatial regularization and the Markov diffusion process. The diffusion distance of the middle pixel v.

[0050] Step 4.1.21, through calculation Each unlabeled pixel With the center pixel of each clustering pattern diffusion distance , Minimum value corresponding to The clustering label is Cluster tags enable the propagation of cluster tags to Unlabeled pixels in, i.e. ,in .

[0051] Step 4.1.22, based on The cluster labels of the middle pixels determine the cluster labels of the remaining pixels in the image, by calculating... Unlabeled clustered pixel and Each labeled cluster tag pixel Mahalanobis distance And then Sort in ascending order F( ) The maximum value corresponding to The clustering label is Cluster labels are used to determine the cluster labels of unlabeled pixels. The representativeness (kernel density) of an unlabeled pixel is weighted by the similarity (represented by the reciprocal of the Mahalanobis distance) between the unlabeled and labeled pixels. The larger the F value, the higher the probability that the unlabeled and labeled pixels belong to the same cluster.

[0052] Step 4.1.23, based on All pixels in the image are clustered with labels. Adjacent pixels with the same cluster label are merged to obtain the clustered post-processed pixels.

[0053] Step 4.1.3, following the steps above, for and After clustering, we obtain and .

[0054] Step S5: Extract change patches from the optimized segmentation and clustering results using the semantic information change detection operator of dual-temporal latent space matching, obtain change patches, and calculate change features.

[0055] Specifically, the semantic information change detection operator based on two-phase potential space matching is used to extract the change patches from the segmentation and clustering optimization results of the pre-period image and the post-period image. and the segmentation and clustering optimization results of the post-period image.

[0056] On this basis, step S5 includes: extracting the potential features of the pre-period image and the post-period image by the image encoder of the SAM model, calculating the average value of the potential features in each channel in each patch of the optimization segmentation and clustering results to obtain a potential feature vector; calculating the patch change confidence score according to the potential feature vector to obtain the change degree of the pre-period image and the post-period image; obtaining a patch change confidence score threshold, and in response to the patch change confidence score being greater than the patch change confidence score threshold, taking the patch as the extracted change patch.

[0057] calculating the patch change confidence score according to the potential feature vector to obtain the change degree of the pre-period image and the post-period image includes: calculating the patch change confidence score according to the following formula: ; ; wherein, the patch change confidence score, the patch change confidence score of any patch of the pre-period image, the patch change confidence score of any patch of the post-period image, the patch change confidence score of any patch, the patch serial number, the potential feature vectors of the corresponding patches at the same position of the pre-period image and the post-period image, respectively.

[0058] Specifically, step 5.1 extracts the potential features of the pre-period image and the post-period image by using the encoder of the open-source SAM model 、 calculates the average value of in each patch object of as the pre-period potential representation vector of the patch calculates the average value of in each patch object of as the post-period potential representation vector of the patch .

[0059] Step 5.2 calculates the cosine distance between and ​similarity between the two images, and a change score is calculated to evaluate the semantic change degree of the image patches in the two images. ; where index is the patch number, is the latent representation vector of the patch corresponding to the same position in the two images, The value range of ChangeScore is 0~1, and the higher the value, the greater the semantic change degree of the patch. The above formula is the complement of cosine similarity (Cosine Similarity), which is used to measure the directional difference of the latent feature vectors of the patches in the two images. When the vector directions are completely consistent, the cosine value = 1, = 0 (no change); when the vector directions are orthogonal, the cosine value = 0, = 1 (maximum change).

[0060] Step 5.3, a bidirectional matching change detection strategy is designed to calculate the change confidence score, ensuring the time symmetry of change detection. First, taking the patch of as the reference, the change confidence score of the patch is calculated , then taking the patch of as the reference, the change confidence score of the patch is calculated , and the final patch change confidence score = is set. The patch change confidence score threshold is set. If the threshold is exceeded , it indicates that the patch is a change patch.

[0061] Step S6, calculate the Pearson correlation coefficient according to the change patch and the change feature, and obtain the patch change type.

[0062] On this basis, step S6 includes: Calculate the Pearson correlation coefficient according to the following formula: ; wherein, Pearson correlation coefficient, covariance, feature vector of the indexth change patch, variance of the change feature of the CTth change type; obtain a correlation coefficient threshold, and define the type of the change patch as the corresponding change type corresponding to the corresponding change feature in response to the Pearson correlation coefficient being greater than the correlation coefficient threshold.

[0063] Specifically, the pixel-level change feature and the object-level change feature are calculated according to the calculated index-th change map, so as to obtain a feature vector of the index-th change map , and the Pearson correlation coefficient is calculated with each change type change feature vector , and if the Pearson correlation coefficient exceeds a threshold PT, it is indicated that the change type of the change map index is the CT-th change type. ; Among them, is the covariance, , is the variance of .

[0064] The change map image is converted into a vector, and an attribute field "change type" is added, the belonging change type of each change map is determined according to the Pearson correlation coefficient and the threshold, the belonging change type is assigned to the attribute information of the corresponding change map, the attribute information value is cultivated land changed into economic crops, cultivated land changed into forest land, cultivated land changed into nursery, cultivated land changed into grassland, cultivated land changed into pit pond, cultivated land changed into building, and cultivated land changed into excavation land, so as to obtain a cultivated land non-cerealization remote sensing intelligent monitoring result vector.

[0065] The mountainous cultivated land non-cerealization remote sensing intelligent monitoring method provided by the application has the following advantages: (1) Based on the results of semantic segmentation of the pre-period and post-period high-resolution remote sensing images by the open source semantic segmentation model SAM, a zero sample cultivated land non-cerealization change detection method including two key steps of spatial regularization diffusion learning clustering algorithm and double-phase latent space matching semantic information change detection operator is designed, the change map is found out by using the zero sample change detection strategy, and then the features such as color, texture and shape of the pre-image and post-image in the change map are extracted, and the similarity is determined with the cultivated land non-cerealization change detection sample library, and then the change map of the cultivated land non-cerealization is filtered out, so as to provide a new solution for the fine recognition of cultivated land, the prevention of cultivated land "non-cerealization" and the overall consolidation of food security. (2) The method only needs a small amount of manually labeled typical samples of cultivated land non-cerealization change region, deeply mines the feature differences between the pre-period and post-period high-resolution images, avoids large-scale labeling of non-cerealization change region interpretation samples and training of the model, saves manpower and material resources, and has the advantages of lower cost, higher efficiency and better performance compared with the full-supervised deep learning semantic segmentation algorithm.

[0066] It should be noted that the method of the embodiments of the present application can be executed by a single device, such as a computer or a server, etc. The method of the embodiments can also be applied to a distributed scenario, and be completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only execute one or more steps in the method of the embodiments of the present application, and the multiple devices can interact with each other to complete the method.

[0067] It should be noted that some embodiments of the present application have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than that described above and still achieve desirable results. Additionally, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0068] Based on the same inventive concept, the present application also provides a mountainous cultivated land non-food remote sensing intelligent monitoring system for zero sample change detection corresponding to the method of any of the above embodiments.

[0069] Reference Figure 3 The mountainous cultivated land non-food remote sensing intelligent monitoring system for zero sample change detection comprises: A data acquisition module 201 is configured to acquire high-resolution optical images of a to-be-detected region in different years, cultivated land survey and monitoring vectors corresponding to the years, and economic crop survey data of the following year; perform geometric precise correction and pixel-level position registration on the high-resolution optical images to obtain image data of the previous period and the later period; A change sample acquisition module 202 is configured to acquire change characteristics of each change type in a cultivated land non-food change detection typical sample library; wherein the change types include: cultivated land changing into economic crops, cultivated land changing into forest land, cultivated land changing into nursery, cultivated land changing into grassland, cultivated land changing into pit pond, cultivated land changing into building, and cultivated land changing into excavation land; An image segmentation module 203 is configured to perform sliding cropping on the image data of the previous period and the later period to obtain an image small piece group; wherein the image small piece group comprises one previous period image small piece and one later period image small piece; segment the image small piece group according to a SAM model and perform small piece splicing after segmentation to obtain a segmentation result of the image; A clustering optimization module 204 is configured to perform clustering post-processing on the segmentation result of the image by a spatial regularization diffusion learning clustering algorithm to obtain an optimized segmentation clustering result; The change extraction module 205 is configured to extract change patches from the optimized segmentation clustering result by a semantic information change detection operator based on dual-time potential space matching, to obtain change patches and calculate change features. The type acquisition module 206 is configured to calculate a Pearson correlation coefficient according to the change patches, the change features, and the change features of each change type in the typical sample library, to obtain patch change types.

[0070] For the convenience of description, the above system is described in various modules in terms of functions. Of course, the functions of the modules can be implemented in one or more software and / or hardware when implementing the present application.

[0071] The system of the above embodiment is used to implement the mountainous cultivated land non-food remote sensing intelligent monitoring method of zero sample change detection in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be described here.

[0072] Those skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope (including claims) of the present application is limited to these examples; under the idea of the present application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes of different aspects of the embodiments of the present application as described above. In order to be brief, they are not provided in detail.

[0073] Any process or method descriptions in flow charts or otherwise described herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for implementing specific logic functions (or steps) in the process, and the various embodiments of the present application include additional implementations in which the order of steps can be different, including use of simultaneous processes, or the steps can be performed in reverse order, or at least some steps can be performed concurrently, or with prior processes or other steps, depending upon the functionality involved. This can be understood by those skilled in the art.

[0074] Having described specific embodiments of the application with reference to the accompanying drawings, it is apparent that the embodiments of the application can be implemented without such specific details or with variations on the specific details. Therefore, these descriptions should be considered illustrative rather than restrictive. Although the present application has been described in connection with specific embodiments thereof, many alternatives, modifications and variations will be apparent to those skilled in the art in light of the foregoing description.

[0075] Embodiments of the present application are intended to cover all such alternatives, modifications, and variations as fall within the broad scope of the appended claims. Accordingly, any one or more features of any embodiment of the present application can be included in, or prior art to, any one or more features of any other embodiments of the present application; and no characteristics mentioned in this specification should be considered essential (i.e., interpreted in the context of a "means-plus-function" or "step-plus-function" clause) unless the specification expressly states the characteristic is essential.

Claims

1. A method for intelligent remote sensing monitoring of non-grainization of mountainous farmland without sample change detection, characterized in that, The method comprises the following steps: Step S1, obtaining high-resolution optical images of a to-be-detected area in different years, cultivated land survey monitoring vectors corresponding to the years, and economic crop survey data of the next year; Performing geometric precise correction and pixel-level position registration on the high-resolution optical images to obtain image data of the early stage and the later stage; Step S2, obtaining change characteristics of each change type in a cultivated land non-food crop change detection typical sample library; wherein the change types include: cultivated land changing into economic crops, cultivated land changing into forest land, cultivated land changing into nursery, cultivated land changing into grassland, cultivated land changing into pit pond, cultivated land changing into building, and cultivated land changing into excavation land; Step S3, performing sliding cutting on the image data of the early stage and the later stage to obtain an image small piece group; wherein the image small piece group comprises one early-stage image small piece and one later-stage image small piece; performing segmentation on the image small piece group according to a SAM model and performing small piece splicing after segmentation to obtain a segmentation result of the image; Step S4, performing clustering post-processing on the segmentation result of the image by a spatial regularization diffusion learning algorithm to obtain an optimized segmentation clustering result; Step S5, performing change patch extraction on the optimized segmentation clustering result by a semantic information change detection operator of a dual-time-phase latent space matching to obtain change patches and calculate change characteristics; Step S6, calculating a Pearson correlation coefficient according to the extracted change patches and change characteristics and change characteristics of each change type in the typical sample library to obtain a patch change type.

2. The method according to claim 1, wherein, The method further comprises the following steps before step S2: Filtering out cultivated land changing into non-cultivated land patches from the cultivated land survey monitoring vectors and the economic crop survey data by using overlay analysis, obtaining typical cultivated land changing into non-cultivated land patches, and making a cultivated land non-food crop change detection typical sample library; Obtaining a cultivated land non-food crop change detection typical sample library and calculating change characteristics of each change type in the cultivated land non-food crop change detection typical sample library.

3. The method according to claim 2, wherein, The step of obtaining a cultivated land non-food crop change detection typical sample library and calculating change characteristics of each change type in the cultivated land non-food crop change detection typical sample library comprises the following steps: Calculating change characteristics of a typical sample patch in a change type, wherein the change characteristics of the typical sample patch comprise pixel-level change characteristics and object-level change characteristics; ； ； ； ； ； ； ； ； ； ； wherein, represent the first typical example image patch pixel-level change characteristics, respectively, are the first typical example image patch difference between the mean values of the red, green and blue band histograms in the pre- and post-stage image data, respectively, are the first typical example image patch difference between the standard deviations of the red, green and blue band histograms in the pre- and post-stage image data, represent the first typical example image patch object-level change characteristics, respectively, are the first typical example image patch difference between the roughness, contrast, directionality, linearity, regularity, coarseness texture characteristics of the red, green and blue bands in the pre- and post-stage image data, respectively, are the first typical example image patch difference between the morphological top-hat transform characteristics and the morphological bottom-hat transform characteristics of the red, green and blue bands in the pre- and post-stage image data, respectively, represent the first typical example image patch difference between the roughness of the red, green and blue bands, respectively, represent the first typical example image patch difference between the contrast of the red, green and blue bands, respectively, represent the first typical example image patch difference between the directionality of the red, green and blue bands, respectively, represent the first typical example image patch difference between the morphological top-hat transform characteristics of the red, green and blue bands, and The change characteristics of the typical sample patch obtain change characteristics of the change type; ； ； ； in, This represents the average value of pixel-level variation features. This represents the number of all typical sample patches under the CT-type variation. Indicates the first A typical example image patch, Indicates object-level change characteristics. This represents the change characteristics of the CT-th type of change.

4. The method of claim 1, wherein the method is characterized by, The step S4 comprises the following steps: Calculating kernel density according to the segmentation result of the image, selecting representative pixels according to a preset kernel density requirement, constructing a spatial regularization KNN neighbor connection graph according to the representative pixels, and obtaining a reduced spatial regularization diffusion graph; Locating K pixels as clustering mode centers according to the reduced spatial regularization diffusion graph, propagating clustering labels to unmarked pixels in the reduced spatial regularization diffusion graph, determining clustering labels of the remaining pixels according to the clustering labels of the pixels in the reduced spatial regularization diffusion graph, and obtaining an optimized segmentation clustering result.

5. The method of claim 4, wherein the method further comprises: The step of calculating kernel density according to the segmentation result of the image, selecting representative pixels according to a preset kernel density requirement, constructing a spatial regularization KNN neighbor connection graph according to the representative pixels, and obtaining a reduced spatial regularization diffusion graph comprises the following steps: Calculating kernel density according to the following formula: ； wherein, denotes the kernel density of pixels within each map object, denotes the th pixel in , the set of pixels with the closest Euclidean distance to the pixel in the location of the segmentation result within a radius of R pixels, , respectively denote the pixel value of pixel in the red band of the image, , respectively denote the pixel value of pixel in the green band of the image, , respectively denote the pixel value of pixel in the blue band of the image, denotes a scaling factor for controlling the interaction radius between pixels, denotes a regularization factor, ensures ;​ In each graph spot object, kernel density of each pixel is arranged in descending order, and the first k pixels are selected as representative pixels of the graph spot object; The spatial regularization KNN neighbor connection graph is constructed according to the following formula to obtain a reduced spatial regularization diffusion graph: ； wherein, denotes a reduced spatially regularized diffusion map, denotes the total number of patches, denotes the th patch, the number of pixels in .

6. The method of claim 5, wherein the method further comprises: The K pixels are positioned as clustering mode centers according to the reduced spatial regularization diffusion graph, the clustering labels are propagated to unmarked pixels in the reduced spatial regularization diffusion graph, the clustering labels of the remaining pixels are determined according to the clustering labels of the pixels in the reduced spatial regularization diffusion graph, and an optimized segmentation clustering result is obtained, including: The clustering mode center representation values are calculated according to the following formula, and the K pixels with the largest clustering mode center representation values are selected as the clustering mode centers: ； ； wherein, denotes a cluster mode center representation value, denotes a pixel at time has a higher density of diffusion distances between nearest neighbors, time for the diffusion map and Markov diffusion process according to the spatially reduced regularization, denotes a pixel has a higher density of pixels in the diffusion map, denotes a spatially reduced regularization diffusion map; The clustering labels are propagated to unmarked pixels in the reduced spatial regularization diffusion graph according to the following formula: ； ； wherein, denotes the cluster label function of pixel X, denotes the unlabelled pixels in the spatially regularized diffusion map, denotes the distance the nearest cluster center pixel, denotes the parameter that minimizes the function value, denotes the spatially regularized diffusion map, denotes the cluster mode center, the values of are 1, 2, 3,... K, denotes the pixel and the diffusion distance at time ; The correlation degree values are calculated according to the following formula, and the clustering label of the pixel with the largest correlation degree value is selected: ； wherein, denotes the degree of association value, denotes the unmarked cluster label pixel with the reduced spatial regularized diffusion map Mahalanobis distance, of each marked cluster label pixel, denotes the pixel kernel density estimate value itself; Adjacent graph spots with the same clustering label are merged to obtain the optimized segmentation clustering result.

7. The mountainous cultivated land non-food remote sensing intelligent monitoring method of zero sample change detection according to claim 1, characterized in that, The step S5 includes: Potential features of the early image and the late image are extracted by an image encoder of the SAM model, and the potential features in each channel are averaged in each graph spot of the optimized segmentation clustering result to obtain a potential feature vector; A graph spot change confidence score is calculated according to the potential feature vector to obtain a change degree of the graph spot semantics of the early image and the late image; A graph spot change confidence score threshold is obtained, and in response to the graph spot change confidence score being greater than the graph spot change confidence score threshold, the graph spot is taken as an extracted change graph spot.

8. The method according to claim 7, wherein, The graph spot change confidence score is calculated according to the potential feature vector to obtain a change degree of the graph spot semantics of the early image and the late image, including: The graph spot change confidence score is calculated according to the following formula: ； ； wherein, denotes a confidence score of a region change, denotes a confidence score of a region change of any region of the pre-stage image tile, denotes a confidence score of a region change of any region of the post-stage image tile, denotes a confidence score of a region change of any region, denotes a region number, respectively denote latent feature vectors of corresponding regions at the same location of the pre-stage image and the post-stage image.

9. The method of claim 1, wherein the method is characterized by, The step S6 includes: The Pearson correlation coefficient is calculated according to the following formula: ； wherein, denotes the Pearson correlation coefficient, denotes the covariance, denotes the eigenvector of the indexth variation tile, denotes the variance of the variation feature of the CTth variation type; A correlation coefficient threshold is obtained, and in response to the Pearson correlation coefficient being greater than the correlation coefficient threshold, the type of the change graph spot is defined as a change type corresponding to the change feature.

10. A remote sensing intelligent monitoring system for non-grainization of mountainous farmland for zero sample change detection, characterized in that, A mountainous cultivated land non-food remote sensing intelligent monitoring method for realizing zero sample change detection according to any one of claims 1-9, comprising: A data acquisition module is configured to acquire high-resolution optical images of a to-be-detected region in different years, cultivated land survey and monitoring vectors corresponding to the years, and economic crop survey data in the next year; geometric precision correction and pixel-level position registration are performed on the high-resolution optical images to obtain early and late image data; A change sample acquisition module is configured to acquire change features of each change type in a cultivated land non-food change detection typical sample library; wherein the change types include: cultivated land changing into economic crops, cultivated land changing into forest land, cultivated land changing into nursery, cultivated land changing into grassland, cultivated land changing into pit pond, cultivated land changing into building, and cultivated land changing into excavation land; An image segmentation module is configured to perform sliding cropping on the early and late image data to obtain image small piece groups; wherein the image small piece groups include one early image small piece and one late image small piece; the image small piece groups are segmented according to a SAM model and segmented small pieces are spliced to obtain a segmentation result of the image; The clustering optimization module is configured to perform post-processing on the segmentation result of the image by using a spatial regularization diffusion learning clustering algorithm to obtain an optimized segmentation clustering result. The change extraction module is configured to perform change patch extraction on the optimized segmentation clustering result by using a semantic information change detection operator based on a dual-time latent space matching to obtain change patches and calculate change features. The type acquisition module is configured to calculate a Pearson correlation coefficient according to the change patches, the change features, and change features of each change type in a typical sample library to obtain patch change types.

Images