Remote sensing driven land utilization monitoring system

Abstract

The invention relates to the technical field of remote sensing image processing, in particular to a remote sensing driven land utilization monitoring system. According to the invention, by extracting a ground feature boundary contour in a remote sensing image and constructing a morphological parameter set with geometric structure proportion characteristics, a path category data system for dividing a land spatial form based on structural characteristics is established, and normalization processing and a Gaussian mixture model are introduced to realize ground feature aspect ratio clustering. Constructing a structural difference index set in combination with the distribution concentration ratio, clearly identifying the path type to track the land parcel direction or the structure change trajectory, and calculating a disturbance factor in the multi-temporal remote sensing image based on the land type number change and the connection direction number of the patch adjacent region; and the main direction, the boundary contour similarity and the spatial disturbance of the land parcel in the specified period are comprehensively judged, so that the dynamic change identification of the land utilization state is realized, and the land utilization form change amplitude and the time period characteristics can be accurately reflected by setting a judgment reference and a change state marking mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image processing, and particularly relates to a remote sensing driven land use monitoring system. BACKGROUND

[0002] The technical field of remote sensing image processing is a technical system for information extraction and understanding of images collected by remote sensing equipment, and its core matters include discrimination of ground targets, quantization of distribution characteristics, and expression and modeling of spatial attributes based on remote sensing imaging data.

[0003] Among them, the traditional remote sensing driven land use monitoring system refers to a system for analyzing and interpreting the land use form based on remote sensing image data, which is used to identify land use types from remote sensing images and monitor the spatial pattern changes.

[0004] The traditional remote sensing driven land use monitoring system mainly relies on the analysis and interpretation process of ground object types in remote sensing images, and the boundary form structure of ground objects is not refined in the identification process, which makes it difficult to accurately distinguish different ground object categories in areas where there are multiple similar forms but significant differences in use, and it is difficult to capture the changes in land state in time when the spatial pattern is disturbed in microstructure due to the lack of dynamic change correlation judgment mechanism of adjacent plots. When the multi-period remote sensing image shows slight boundary changes or the evolution trend of the land class is not obvious, the system may miss the change due to the lack of continuity parameter comparison mechanism, and it also lacks further classification ability for the identified changed ground objects, making it difficult to accurately reflect the specific type and land evolution trend of the changes. For example, forest land and shrubs are easily misjudged due to similar forms, or construction land is misidentified due to overlapping spectral characteristics with farmland in the early stage of conversion, resulting in certain limitations in the timeliness and classification accuracy of the land use monitoring results. SUMMARY

[0005] The purpose of the present application is to solve the problems existing in the prior art and to provide a remote sensing driven land use monitoring system.

[0006] In order to achieve the above purpose, the present application adopts the following technical scheme: a remote sensing driven land use monitoring system, the system comprises: A patch form extraction module extracts the patch boundary contour of the ground object on the land surface from the remote sensing image, and constructs a patch form parameter set by proportionally calculating the geometric structure characteristics of the patch boundary contour. A path type division module divides the land surface form structure according to the patch form parameter set, sets the path type identifier according to the form structure, and forms a path category data set. An adjacent change identification module judges the land class change of each patch adjacent area based on the path category data set through the remote sensing image, and calculates the land class disturbance factor of the corresponding patch. Land change monitoring module: according to the path type specified in the path category data set of each patch, compared with the morphological parameter change of each patch in the specified period, judge whether the land use state changes, get the land use change monitoring result.

[0007] The patch morphological parameter set includes aspect ratio, boundary direction and spatial distribution position, the path category data set includes path type, morphological zoning category and patch path mapping relationship, the land class disturbance factor includes adjacent land class number change, adjacent direction number and change frequency statistical value, and the land use change monitoring result includes change mark, morphological change amplitude and judgment time period.

[0008] The patch morphological extraction module comprises: Boundary recognition submodule: segmenting the land use object patch in the remote sensing image, collecting the boundary line pixel position of each patch in the remote sensing image, calculating the coordinate point set of the boundary line in the remote sensing image grid, and generating the boundary line coordinate data set; Structure operation submodule: based on the boundary line coordinate data set, extracting the outermost side length data of the envelope frame of each patch boundary, judging the direction attribute of the two sides of the envelope frame and calculating the length proportion of the corresponding direction, and generating the boundary structure ratio sequence; Parameter collection submodule: according to the boundary structure ratio sequence and the patch position index in the remote sensing image coordinate system, integrating the number, position coordinate and geometric ratio of each patch, and obtaining the patch morphological parameter set.

[0009] The path type division module comprises: Cluster recognition submodule: calling the aspect ratio value of each patch in the patch morphological parameter set, taking the normalized aspect ratio value as the input variable, and adopting Gaussian mixture model algorithm to cluster all patches to obtain patch clustering distribution result; Regional classification submodule: based on the mean value and covariance of the aspect ratio of each cluster type in the patch clustering distribution result, calculating the numerical deviation degree of each type of patch on the aspect ratio feature, and comparing the distribution concentration degree of each type of patch on the remote sensing image, and generating the patch structure difference index set; Path setting submodule: according to the combination features of each type of patch on the aspect ratio and spatial concentration degree in the patch structure difference index set, setting two types of path type identification of structure recognition path and direction adjustment path, pairing the path type and patch number to establish the corresponding relationship, and obtaining the path category data set.

[0010] The adjacent change recognition module comprises: Adjacent extraction submodule: obtain remote sensing images at multiple time points, call the spatial position of each patch in the path category dataset, identify adjacent land object patches connected by the boundary, and extract the number of adjacent patches to obtain an adjacent patch index set; Land class change determination submodule: based on the adjacent patch index set, collect the land class numbers of adjacent patches at two remote sensing image time points, compare the number change, and summarize the number of patches with changes to obtain an adjacent land class change count result; Disturbance quantity calculation submodule: call the adjacent land class change count result, combine the number of directional connections between each patch boundary and adjacent patches in the current time point, calculate the correlation quantity of the number of adjacent patch land class changes and the number of connection directions, and obtain the land class disturbance factor corresponding to each patch.

[0011] The land change monitoring module comprises: Path calling submodule: call the path type corresponding to each patch in the path category dataset, identify the directional adjustment path or the structure recognition path, obtain the main direction angle or the boundary contour similarity value of the patch in two remote sensing image periods according to the path type, and obtain a path type parameter set; Parameter comparison submodule: based on the path type parameter set, combine the land class disturbance factor of each patch, set a form parameter change judgment reference matched with the path type, calculate the parameter change amplitude of each patch in two periods, and generate a patch change comparison quantity; Change judgment submodule: compare the patch change comparison quantity with the set judgment reference, assign change state marks and stable state marks to the patch according to the comparison result, and obtain a land use change monitoring result.

[0012] The land use change monitoring result comprises: The change classification result comprises cultivated land change types, forest land change types, water body change types, and construction land change types.

[0013] The change type discrimination module comprises: Patch screening submodule: extract the patch number with a change state mark in the land use change monitoring result, and determine the image positioning information in the current period remote sensing image to obtain a change patch index set; Feature extraction submodule: based on the change patch index set, obtain the principal component feature data and spectral reflection information of the corresponding patch in the current period remote sensing image, and collect the land class number corresponding to the patch in the last period to form change patch feature combination data; The type determination sub-module: according to the category difference between the current land type and the historical land type, the spectrum value change direction and the spatial structure change amplitude in the change patch feature combination data, the change patch is divided into multiple categories, and the change classification result is obtained.

[0014] Compared with the prior art, the advantages and positive effects of the present application are that: In the present application, by extracting the feature boundary contour in the remote sensing image and constructing a morphological parameter set with geometric structure proportion characteristics, a path category data system for dividing land spatial form based on structure characteristics is established, the normalized processing and Gaussian mixture model are introduced to realize the aspect ratio clustering of features, the structure difference index set is constructed in combination with the distribution concentration degree, the path type is clearly identified to track the land block direction or structure change track, the disturbance factor is calculated based on the land type number change and the connection direction number of the patch adjacent area in the multi-temporal remote sensing image, and the main direction, the boundary contour similarity and the spatial disturbance of the land block in the specified period are comprehensively judged based on the disturbance factor, so that the dynamic change recognition of the land use state is realized, the change amplitude and the time period characteristics of the land use form are accurately reflected by setting the judgment reference and the change state marking mechanism, and after the change recognition, the fine classification of the land use change type is realized based on the principal component feature, the spectral reflection and the historical land type number, etc., so that the fine-grained perception and classification judgment ability for the change of the cultivated land, the forest land, the water body and the construction land is effectively enhanced, and the ability of the remote sensing monitoring system for the structured modeling, the continuous evolution perception and the change type identification of the land use evolution is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The system module diagram of the present application is shown in the figure; Figure 2 The system framework diagram of the present application is shown in the figure; Figure 3 The schematic diagram of the patch form extraction module of the present application is shown in the figure; Figure 4 The schematic diagram of the path type division module of the present application is shown in the figure; Figure 5 The schematic diagram of the adjacent change recognition module of the present application is shown in the figure; Figure 6 The schematic diagram of the land change monitoring module of the present application is shown in the figure; Figure 7 The schematic diagram of the change type discrimination module of the present application is shown in the figure. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the figures and examples. It should be understood that the specific examples described herein are only used to explain the present application, and are not used to limit the present application.

[0017] In the description of the present application, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0018] Referring to Figure 1 The present application provides a technical solution: a remote sensing driven land use monitoring system, the system comprising: Patch morphology extraction module: extracting the land surface feature patch boundary contour from the remote sensing image, constructing the patch morphology parameter set by proportionally calculating the geometric structure characteristics of the patch boundary contour; Path type division module: referring to the patch morphology parameter set to divide the land surface morphology structure, setting the path type identifier according to the morphology structure, and forming the path category data set; Adjacent change identification module: based on the path category data set, judging the land class change of each patch adjacent area through the remote sensing image, and calculating the land class disturbance factor of the corresponding patch; Land change monitoring module: according to the path type specified in the path category data set for each patch, comparing the morphology parameter change of each patch in the specified period with the land class disturbance factor of each patch, judging whether the land use state has changed, and obtaining the land use change monitoring result; The patch morphology parameter set includes aspect ratio, boundary direction, spatial distribution position, the path category data set includes path type, morphology zoning category, patch path mapping relationship, the land class disturbance factor includes adjacent land class number change, adjacent direction number, change frequency statistical value, and the land use change monitoring result includes change mark, morphology change amplitude and judgment time period.

[0019] Referring to Figure 2 And Figure 3 The patch morphology extraction module comprises: Boundary identification submodule: segmenting the land use feature patch in the remote sensing image, collecting the boundary line pixel position of each patch in the remote sensing image, calculating the coordinate point set of the boundary line in the remote sensing image grid, and generating the boundary line coordinate data set; First, the remote sensing image is acquired, and an image segmentation algorithm such as U-Net is used to classify and process the remote sensing image, and each type of ground feature in the image such as water, buildings, farmland, forest land, etc. is labeled as an independent patch. Then, an edge detection algorithm such as Canny operator or Sobel operator is used to obtain the boundary pixel points of each patch. For example, in a 512x512 pixel remote sensing image, after identifying an independent patch with an area greater than 100 pixels, the Canny edge detection algorithm is used to extract the contour points of one of the farmland patches, a total of 300 boundary pixel points are obtained. The image coordinates (row and column numbers) of these points are converted into actual positions in the geographic coordinate system or map projection coordinate system. For example, using the WGS84 coordinate system, the geographic coordinates of the boundary points can be obtained by coordinate transformation based on the affine transformation parameters of the remote sensing image. For example, a boundary point with a pixel position of (250, 130) is converted to geographic coordinates (113.4567, 34.1234) after affine transformation. Repeat the process to obtain the complete boundary line coordinate sequence and form the boundary line coordinate data set.

[0020] Structure operator module: based on the boundary line coordinate data set, the outermost edge length data of the envelope box of each patch boundary is extracted, the direction attribute of the envelope box is judged, and the length ratio of the corresponding direction is calculated to generate a boundary structure ratio sequence; Based on the boundary line coordinate data set, the boundary structure features of each patch are processed. First, the minimum circumscribed rectangle method is applied to each group of boundary line coordinates to extract the envelope box edge length information. This algorithm determines the length and width of the rectangular box by calculating the maximum and minimum horizontal and vertical coordinates of the boundary points. For example, the boundary line horizontal coordinate range of a patch is 113.45-113.55, with a latitude and longitude span of 0.10, which is equivalent to 111 kilometers at 1 degree. The horizontal edge length is 11.1 km, and the vertical coordinate range is 34.10-34.16, with a vertical edge length of 6.66 km. The direction attribute is determined by assigning a direction label (such as horizontal) based on the direction of the long side (latitude or longitude). Then, the length-width ratio of the envelope box is calculated as 11.1:6.66=1.67, which is the boundary structure ratio. This value is recorded in the sequence for subsequent comparison and classification. If another patch calculates the envelope box ratio as 0.95, it belongs to the near-square category. By iterating through all patches and performing this type of ratio extraction, multiple boundary structure ratio sequence sets are formed.

[0021] Parameter collection submodule: based on the boundary structure ratio sequence and the patch position index in the remote sensing image coordinate system, the number, position coordinates, and geometric ratio of each patch are integrated to obtain a patch morphology parameter set; Based on the sequence of boundary structure ratio, combined with the position information of the patch in the remote sensing image, the morphological parameter set of each patch is sorted. Firstly, the boundary structure ratio corresponding to each number is extracted by the number of each patch in the image, such as labeling number 1~N according to the connected region in the segmented image, and the geometric center coordinates of the patch are obtained from the boundary coordinate set of the patch, for example, the boundary center coordinates of a patch with number 7 are (113.4789, 34.1432), and the corresponding structure ratio is 1.25, which constitutes the mapping relationship between patch number and position, and further integrates the patch number, center coordinates and boundary structure ratio into a morphological parameter set, such as number 7-(113.4789, 34.1432)-1.25, and the operation is performed on all patches to form a batch parameter set, which can be used for structure statistics and distribution analysis of ground objects, and finally the morphological parameter set of the patch is obtained.

[0022] Please refer to Figure 2 and Figure 4 , the path type division module comprises: The cluster identification submodule calls the aspect ratio value of each patch in the patch morphological parameter set, and the normalized aspect ratio value is used as an input variable. The Gaussian mixture model algorithm is used to cluster all patches to obtain the patch clustering distribution result. After completing the normalization of the aspect ratio of the patch, the normalized result is continued to be used as an input, and a Gaussian mixture model (GMM) is used for clustering analysis. The clustering target of this time is to divide all patches into three categories according to the normalized aspect ratio. The core model of GMM is the weighted superposition of multiple Gaussian distributions, and the total probability density function is: ; According to the maximum probability principle, the value of the total probability density function under each Gaussian distribution category (such as low, medium and high aspect ratio categories) is compared, and the patch is classified into the category with the maximum probability value; wherein, : input variable, i.e. normalized aspect ratio of the patch, value range is [0, 1]; : the number of set clustering categories, in this example, it is 3, i.e. ; : the weight of the first category, which satisfies , indicating the proportion of samples belonging to the category, and the weight is proportional to the number of sample points in the category, for example, the more sample points in the category, the greater the weight; : the mean of the first category, indicating the center value of the normalized aspect ratio of the category; : the variance of the first category, reflecting the dispersion degree of the data in the category; : taking as the mean, The normal distribution function for variance.

[0023] Suppose there are 6 patches with normalized aspect ratio values: 0.15, 0.18, 0.62, 0.68, 0.89, 0.93.

[0024] In the initial step, the number of clusters is set , and the parameters of each cluster are initialized as: Cluster 1 (low aspect ratio cluster): , , ; Cluster 2 (medium aspect ratio cluster): , , ; Cluster 3 (high aspect ratio cluster): , , .

[0025] For a certain input data , the probability of belonging to each cluster is calculated as follows: Calculate the probability term belonging to Cluster 1: ; (The exponential term tends to 0 due to too far distance) Calculate the probability term belonging to Cluster 2: ; Calculate the probability term belonging to Cluster 3: .

[0026] Normalize the above results to get the posterior probability, finally is classified into Cluster 2, indicating that it is a medium aspect ratio cluster patch.

[0027] This process is repeated for all samples, the E step is used to update the probability value of each patch belonging to each cluster, and the M step updates , , , for example, if Cluster 2 contains 3 patches in the next round, its weight is updated to , and the mean and variance are recalculated according to the samples. After convergence, each patch will be uniquely classified into a cluster, and the final clustering distribution label sequence will be formed.

[0028] Specifically, for each patch sample, input its normalized aspect ratio value , then use the probability density function of Gaussian mixture model to calculate its probability value under each Gaussian distribution component, i.e. calculate its probability density under each class (e.g. Cluster 1, Cluster 2, Cluster 3) multiplied by the mixing weight of that class These values essentially represent the likelihood of the patch belonging to each class.

[0029] Subsequently, the weighted probability of each patch under different classes is normalized so that the sum of probabilities of all classes is 1, forming the attribution probability of each class. Specifically, for a certain patch, if its probability under class 1 is 0.1, under class 2 is 0.7, and under class 3 is 0.2, then after normalization, the probability of belonging to class 1 is 0.1 / (0.1+0.7+0.2)=0.1, class 2 is 0.7, and class 3 is 0.2.

[0030] Next, according to these attribution probabilities, the "maximum a posteriori principle" is used for class determination, i.e. the patch is divided into the class with the highest attribution probability. In the above example, the patch has the highest attribution probability to class 2, so it is assigned to class 2.

[0031] This process is repeated for all patches, i.e. the normalized aspect ratio is input one by one, the weighted probability density under each class Gaussian distribution is calculated, the attribution probability of each class is normalized, and then the class to which it belongs is determined according to the maximum probability. Finally, all patches are assigned to a specific class, forming a complete patch clustering distribution result.

[0032] Region classification submodule: based on the mean and covariance of the aspect ratio of each cluster type in the patch clustering distribution result, the numerical deviation of each class of patch in the aspect ratio feature is calculated, and the distribution concentration of each class of patch in the remote sensing image is compared to generate a set of patch structure difference indicators; After obtaining the clustering category of each patch, the average aspect ratio and covariance of each category are counted as the basis for feature analysis. There are three clustering categories, namely low aspect ratio category, medium aspect ratio category and high aspect ratio category. The average aspect ratio and covariance of each category are counted and calculated, respectively. The average of the low aspect ratio category is 0.92, and the covariance is 0.0047. The average of the medium aspect ratio category is 1.08, and the covariance is 0.0012. The average of the high aspect ratio category is 1.26, and the covariance is 0.0065. Then, the difference between the aspect ratio of each patch and the average aspect ratio of the category to which it belongs is used as the evaluation basis for the numerical deviation degree. For example, if the aspect ratio of a patch is 1.00 and it belongs to the medium aspect ratio category, the difference between 1.00 and the average 1.08 of the category is 0.08, which is less than 0.10 and is determined as “mild deviation”. If the aspect ratio of a patch is 0.85 and it belongs to the low aspect ratio category, the difference between 0.85 and the average 0.92 of the category is 0.07, which is in the “moderate deviation” range. If the aspect ratio of a patch is 1.42 and it belongs to the high aspect ratio category, the difference between 1.42 and the average 1.26 of the category is 0.16, which is greater than 0.15 and is determined as “high deviation”. According to the set standard: the difference is less than 0.05, which is “close to the center”; 0.05-0.10 is “mild deviation”; 0.10-0.15 is “moderate deviation”; and greater than 0.15 is “high deviation”. Combined with the spatial distribution concentration of each category of patch on the remote sensing image, such as calculating the average distance between the centers of gravity of each category of patch on the remote sensing image as an evaluation of the concentration degree, less than 300 meters is defined as “high concentration”, 300-600 meters is “medium concentration”, and more than 600 meters is “low concentration”. The deviation degree level and the spatial concentration level are paired to form the structure difference index set of each category of patch, such as “high deviation + low concentration” for the high aspect ratio category, “mild deviation + high concentration” for the medium aspect ratio category, and “moderate deviation + medium concentration” for the low aspect ratio category. The structure difference index set generation is completed.

[0033] Path setting sub-module: according to the combination characteristics of each category of patch in the structure difference index set in terms of aspect ratio and spatial concentration, two types of path types are set, namely structure recognition path and direction adjustment path. The path type and patch number are paired to establish a corresponding relationship, and the path category data set is obtained. Based on the classification results in the structural difference index set, the aspect ratio deviation level and spatial concentration level combination of each class of patches is analyzed, and two types of path types, structure recognition path and direction adjustment path, are divided. The structure recognition path mainly corresponds to the combination of "close to center + high concentration" and "mild deviation + high concentration", which means that the patches have obvious structural characteristics and are concentrated in distribution, and are suitable for direct structure recognition operation. The direction adjustment path corresponds to the combination of "moderate deviation + medium concentration" and "high deviation + low concentration", which means that the patches have obvious fluctuations in morphology or are in a dispersed state in space, and need to be adjusted in direction first, and then structure recognition. For a patch such as No. 18, which belongs to the medium aspect ratio class, the deviation level is "mild deviation" and the concentration level is "high concentration", so the path type is set to structure recognition path. For patch No. 27, which belongs to the high aspect ratio class, the deviation level is "high deviation" and the concentration level is "low concentration", so the path type is set to direction adjustment path. According to this rule, a corresponding relationship between each patch number and path type is established, and finally a path category data set is obtained, which records the numbers of all patches and their matching path type labels.

[0034] Referring to Figure 2 and Figure 5 , the abutment change recognition module comprises: an abutment extraction submodule: obtaining remote sensing images at multiple time points, calling the spatial position of each patch in the path category data set, identifying adjacent land object patches connected by the boundary, and extracting the numbers of adjacent patches to obtain an abutment patch index set; First, prepare two time points of remote sensing image data, such as 2023 and 2025, and call the spatial coordinate information of each patch in the path category data set, determine the actual position of the patch boundary on the image through coordinate mapping, and then use boundary detection to scan the pixel area around the patch boundary. Within a given tolerance range, it is determined whether there is another patch boundary in contact with it, for example, for patch No. 21, it is found that the east side of the boundary is in contact with patch No. 25 for 5 pixels, and the west side is in contact with patch No. 19 for 3 pixels. Then it is determined that No. 21 and No. 25, No. 19 are adjacent patches, and the abutment relationship is recorded. Repeat this process for all patches to establish the number pairing information of each patch and its adjacent patches, and finally form an abutment patch index set, for example, patch No. 21 corresponds to adjacent patches 19 and 25, patch No. 19 corresponds to adjacent patches 21 and 22, and patch No. 25 corresponds to adjacent patches 21 and 27. The entire abutment index set is organized and stored according to patch numbers.

[0035] a land class change determination submodule: based on the abutment patch index set, collecting the land class numbers of adjacent patches at two remote sensing image time points, comparing the number change, summarizing the number of patches with changes, and obtaining the abutment land class change count result; After obtaining the set of adjacent patch indexes, the land class numbers of all patches are read from the two remote sensing image time points respectively, and the land class numbers of adjacent patches at the two time points are compared one by one. For example, if the land class number of patch No. 21 is "farmland" in 2023 and 2025, it is determined that there is no change. If the land class number of patch No. 19 adjacent to patch No. 21 is "forest land" in 2023 and "building" in 2025, it is determined that the land class has changed, and the change is counted as one change event. Continue to check patch No. 25 as "forest land" to "forest land", which is not counted as a change. Only patch No. 19 adjacent to patch No. 21 has changed, and the number of changes is 1. The adjacent land class numbers of all patches are compared, and the number of changes of the adjacent land class of each patch is accumulated. If three of the four adjacent patches of a certain patch number have changed, the number of changes is recorded as 3. The number of changes of the adjacent land class of all patches is summarized to obtain the adjacent land class change count result.

[0036] The disturbance amount calculation submodule calls the adjacent land class change count result, combines the number of directional connections between the boundary of each patch at the current time point and the adjacent patches, calculates the correlation between the number of adjacent patch land class changes and the number of connections, and obtains the land class disturbance factor corresponding to each patch; After obtaining the adjacent land class change count result, the spatial connection relationship between each patch and its adjacent patches at the current time point is combined to calculate the land class disturbance factor of each patch. The factor reflects the strength of the influence of the spatial structure of the area where the patch is located by the change of the adjacent land class. The specific calculation method is as follows: the number of land class changes in all adjacent patches of a certain patch is calculated by ratio with the actual number of connection directions, and the adjacent direction intensity parameter is introduced to optimize the discrimination degree of the structure of the disturbance factor. The unified calculation formula of the disturbance factor is as follows: wherein, : the land class disturbance factor (dimensionless) of patch ; n : the number of land class changes in the adjacent patches of patch (unit: pieces); : the actual number of adjacent directions of patch (unit: pieces), which is usually identified as up, down, left, right and diagonal directions, and the maximum is 8; : the directional connection intensity coefficient, which represents the connection intensity of the adjacent patch with land class change in the direction, and the value range is 0.5-2.0. The longer the boundary contact length, the larger the value. For example, the main boundary contact is more than 10 pixels, which is set to 2.0, 5-10 pixels is 1.5, less than 5 pixels is 1.0, and point contact is 0.5.

[0037] ​There are 5 adjacent directions of patch No. 34 in the current remote sensing image (i.e. ), among which patch No. 31 and patch No. 36 have land class changes in two periods of images, i.e. . The boundary contact pixels between No. 31 and No. 34 are 8, so the direction intensity coefficient is 1.5, and the contact pixels of No. 36 are 3, so the coefficient is 1.0, and the average direction intensity coefficient can be calculated: .

[0038] Finally, the disturbance factor formula is substituted to obtain the disturbance factor of patch No. 34: ; It shows that the patch is moderately affected by the change of adjacent land class at the current time point. The calculation process is performed on all patches to form a complete land class disturbance factor sequence.

[0039] Please refer to Figure 2 and Figure 6 , the land change monitoring module includes: Path calling sub-module: call the path type corresponding to each patch in the path category data set, identify as direction adjustment path or structure recognition path, obtain the main direction angle or boundary contour similarity value of the patch in two remote sensing image periods according to the path type, and obtain the path type parameter set; According to the path type of the patch in the path category data set, the morphological parameters are classified and extracted, if the path type is a structure recognition path, the boundary contour information of the patch in two remote sensing image periods needs to be extracted, and the contour similarity value is calculated. The contour similarity is calculated by using the weighted normalized structure distance model, and its expression is: ; Among them, is the contour similarity of the target patch at two time points, the value range is between 0 and 1, and the closer to 1 indicates that the contour is closer; is the number of comparable boundary points in two time points, is the Euclidean distance between the corresponding boundary points in two time points, the unit is meter, and the normalized processing is , is the average value of the maximum boundary point distance of all patches in the region, the unit is meter; 、 are the boundary contour containing areas of the target patch at two time points, the unit is square meter, and the normalized expression is , is the maximum envelope area in the study area; is the boundary shape complexity weight, which is calculated according to the complexity index , is the boundary length, 0.6 when the index is less than 20, 0.75 between 20 and 30, and 0.9 when greater than 30; 0.4 when the area is less than 2000 square meters, 0.25 between 2000 and 5000, and 0.1 when greater than 5000; the weights reflect that the smaller the area, the more sensitive the change, and the two weights must satisfy: .

[0040] Taking patch No. 52 as an example, 10 comparable boundary points are extracted in two image periods, the Euclidean distances are 12, 8, 9, 7, 11, 10, 9, 8, 6, and 10 meters, the average value between points is 9 meters, the maximum boundary distance normalization factor is 20 meters, and the normalized distance average value is 0.45. The patch boundary circumference is 260 meters, the envelope area is 2300 square meters, the complexity index is , and the two time point areas are set to 2300 and 2100 square meters, respectively, the area difference is 200 square meters, the maximum area is 6400 square meters, the normalized area change is 0.03125, the patch area is in the range of 2000 to 5000 square meters, and the set is 0.25. After substituting into the formula, the calculation is: ; The final contour similarity of patch No. 52 between the two time points is 0.65, indicating that the overall change of the contour is small but there is a certain deformation trend, and this value will be written into the path type parameter set. The basis for similarity judgment is the geometric change degree of the contour between the two time points, including the spatial position difference of the boundary shape and the change amplitude of the envelope area. When the contour similarity is close to 1, the boundary shape of the two time points is highly consistent, the spatial position is basically coincident, and the envelope area changes little; when the similarity is much less than 1, the boundary point is obviously shifted, the contour shape is significantly changed, or the envelope area difference is large. In actual application, the similarity value is compared with the set threshold value, for example, greater than 0.85 can be judged as "highly similar", 0.65 to 0.85 as "moderately similar", and less than 0.65 as "low similarity", which is used to distinguish the stability and change trend of the patch shape over time.

[0041] Parameter comparison submodule: based on the path type parameter set, combined with the land class disturbance factor of each patch, the morphological parameter change judgment benchmark matching the path type is set, the parameter change amplitude of each patch in two periods is calculated, and the patch change comparison quantity is generated; ​After obtaining the path type parameter set and the land class disturbance factor of each patch, different morphological parameter change judgment reference values are set according to the path type. If the path type is a structure recognition path, the contour similarity value is selected for evaluation, and the judgment reference value is set in combination with the disturbance factor. For example, when the disturbance factor is less than 0.3, the similarity reference is set to 0.70, when the disturbance factor is 0.3 to 0.6, the reference is set to 0.75, and when the disturbance factor is higher than 0.6, the reference is set to 0.80, so as to enhance the change sensitivity. If the path type is a direction adjustment path, the main direction angle value is extracted for comparison, and the judgment reference is segmented into 10 degrees, 8 degrees and 5 degrees. The greater the disturbance, the smaller the tolerance. The parameter change amplitude is calculated by taking the absolute value of the difference between the data of two time points. For example, the path type of patch No. 36 is a structure recognition path, the contour similarity is 0.62, the disturbance factor is 0.45, and the corresponding judgment reference is 0.75. Therefore, the change comparison quantity is 0.13, which is the difference between 0.75 and 0.62. The path type of patch No. 41 is a direction adjustment path, the main direction angle at two time points is 85 degrees and 96 degrees respectively, the disturbance factor is 0.28, and the corresponding tolerance angle is 10 degrees. Therefore, the angle change is 11 degrees, which exceeds the reference value. Finally, the change amplitude value of each patch is recorded to form the patch change comparison quantity sequence.

[0042] The change judgment submodule compares the patch change comparison quantity with the set judgment reference value, and according to the comparison result, the patch is marked with a change state mark and a stable state mark to obtain the land use change monitoring result. The patch change comparison quantity is compared with the corresponding judgment reference value. If the change amplitude does not exceed the reference value, the patch is marked with a "stable" state mark. If the change amplitude is equal to or exceeds the reference value, the patch is marked with a "change" state mark. For example, the contour similarity difference value of patch No. 36 is 0.13, and the reference value is 0.75. The actual similarity is lower than the reference value due to the difference, so it is marked as "change". The direction angle difference value of patch No. 41 is 11 degrees, which is higher than the judgment reference 10 degrees, so it is also marked as "change". The direction angle difference of patch No. 55 is 6 degrees, which is lower than the corresponding reference 8 degrees, so it is marked as "stable". After the process is completed for all patches, a patch change state data set is formed, including the number of each patch, the path type, the disturbance factor, the parameter change quantity and the state mark, which is the output result of the land use change monitoring system.

[0043] Please refer to Figure 2 and Figure 7 , which further comprises a change type discrimination module: based on the land use change monitoring result, the land use change type identification is refined to obtain a change classification result; The change classification result includes the change type of cultivated land, the change type of forest land, the change type of water body and the change type of construction land. The change type discrimination module comprises: Patch screening submodule: extract the patch number marked as change in the land use change monitoring result, and determine the image positioning information in the current period remote sensing image to obtain the change patch index set; Filter the land use change monitoring result, extract all patch numbers marked as "change", and further retrieve the spatial positioning information of these patches in the current period remote sensing image, including image row and column number, image coordinates, map projection coordinates, etc., to clearly define the precise range of these patches in the image, such as the boundary start point of patch No. 58 in the image is row and column coordinates (120, 210), and the end point is (145, 235), which can be used for subsequent image data feature extraction. Repeat this positioning process for all patches determined as "change" to finally generate a change patch index set, which is used to identify all areas that need further identification and analysis.

[0044] Feature extraction submodule: based on the change patch index set, obtain the principal component feature data and spectral reflectance information of the corresponding patch in the current period remote sensing image, and collect the land class number corresponding to the patch in the last period to form the change patch feature combination data; With the change patch index set as input, extract the image feature data of each change patch in the current period remote sensing image, mainly including the first few principal component values obtained by principal component transformation and the original spectral reflectance information. For example, patch No. 58 extracts the first principal component value of the current period as 26.5, the reflectance in the green light waveband is 0.34, and the near-infrared waveband is 0.52. At the same time, retrieve the land class number of this patch in the last period as "forest land". Combine the current period image features with the last period land class information to form a complete change patch feature combination data. Process all change patches to obtain a multi-dimensional data set containing image features and historical land class labels.

[0045] Type determination submodule: according to the class difference between the current land class and the historical land class, the spectral value change direction, and the spatial structure change amplitude in the change patch feature combination data, the change patch is divided into multiple categories to obtain the change classification result; Using the changed patch feature combination data, the category difference of each patch current period land class feature and last period history land class, the change direction of spectral reflectance value and the change amplitude of spatial structure are analyzed, the change category is divided through comprehensive judgment, if the last period land class of a patch is "forest land", the current period principal component is significantly decreased, the near infrared band reflectivity is reduced, the boundary contour is simplified, and the spatial structure similarity is less than 0.55, then the patch is judged as "forest land converted to construction land"; if the history land class of a patch is "cultivated land", the current principal component is increased, the red edge reflection is enhanced, and the boundary structure is kept stable, then it is judged as "cultivated land re-cultivated"; if the current principal component amplitude of a patch is not obvious, but the spectral red light segment reflectivity is reduced and the blue light segment is increased, combined with the larger disturbance factor and the medium contour change value, then it is classified as "possible degradation area", after the above judgment is completed for all changed patches, the change classification result set is formed, and the result set can support the regional land use conversion trend statistics and dynamic analysis.

[0046] The above is only the preferred embodiment of the present application, and does not limit the present application in other forms. Any skilled person in the art can use the disclosed technical content to make changes or modifications into equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments without departing from the technical solution content of the present application still belongs to the protection scope of the present application technical solution.

Claims

1. A remote sensing driven land use monitoring system, characterized by: The system comprises: Patch shape extraction module: extract the land surface feature patch boundary contour from the remote sensing image, calculate the geometric structure characteristics of the patch boundary contour by proportion, and construct the patch shape parameter set; Path type division module: refer to the patch shape parameter set to divide the land surface shape structure, set the path type identifier according to the shape structure, and form the path category data set; Adjacent change identification module: based on the path category data set, judge the land class change of each patch adjacent area through the remote sensing image, and calculate the land class disturbance factor of the corresponding patch; Land change monitoring module: according to the path type specified in the path category data set of each patch, compare the morphological parameter change of each patch in the specified period with the land class disturbance factor of each patch, judge whether the land use state changes, and obtain the land use change monitoring result.

2. The remote sensing driven land use monitoring system according to claim 1, characterized in that: The patch shape parameter set includes aspect ratio, boundary direction and spatial distribution position, the path category data set includes path type, morphological zoning category and patch path mapping relationship, the land class disturbance factor includes adjacent land class number change, adjacent direction number and change frequency statistical value, and the land use change monitoring result includes change mark, morphological change amplitude and judgment time period.

3. The remote sensing driven land use monitoring system of claim 1, wherein: The patch shape extraction module comprises: Boundary identification submodule: segment the land use feature patch in the remote sensing image, collect the boundary line pixel position of each patch in the remote sensing image, calculate the coordinate point set of the boundary line in the remote sensing image grid, and generate the boundary line coordinate data set; Structure operation submodule: based on the boundary line coordinate data set, extract the outermost side length data of the envelope frame of each patch boundary, judge the direction attribute of the two sides of the envelope frame and calculate the length proportion of the corresponding direction, and generate the boundary structure ratio sequence; Parameter collection submodule: according to the boundary structure ratio sequence and the patch position index in the remote sensing image coordinate system, integrate the number, position coordinate and geometric ratio of each patch to obtain the patch shape parameter set.

4. The remote sensing driven land use monitoring system of claim 1, wherein: The path type division module comprises: Cluster identification submodule: call the aspect ratio value of each patch in the patch shape parameter set, use the aspect ratio value as the input variable after normalization, use the Gaussian mixture model algorithm to cluster all patches, and obtain the patch clustering distribution result; Regional classification submodule: based on the mean value and covariance of the aspect ratio of each cluster type in the patch clustering distribution result, calculate the numerical deviation degree of each class of patches in the aspect ratio feature, and compare the distribution concentration degree of each class of patches in the remote sensing image, generate the patch structure difference index set; Path setting submodule: according to the combination characteristics of each class of patches in the patch structure difference index set in aspect ratio and spatial concentration degree, set the structure recognition path and the direction adjustment path two path type identifiers, pair the path type with the patch number to establish the corresponding relationship, and obtain the path category data set.

5. The remote sensing driven land use monitoring system of claim 4, wherein: Cluster all patches, use the formula: ； Computing the total probability density function of the Gaussian distributions According to the maximum probability principle, the total probability density function is compared in each Gaussian distribution category, and the plaque is classified into the category with the maximum probability value. wherein, is the set number of clustering categories, is the weight of the th category, representing the proportion of samples belonging to the category, the weight is set in proportion to the number of sample points of the category, is the mean of the th category, also representing the center value of the normalized aspect ratio of the th category, is the variance of the th category, reflecting the dispersion degree of data within the th category, is a normal distribution function with as the mean and as the variance, is the input variable, i.e., the normalized aspect ratio of the patch, with a value range of [0, 1].

6. The remote-sensing driven land use monitoring system of claim 1, wherein: The adjacent change identification module comprises: An adjacent extraction submodule: obtain remote sensing images at multiple time points, call the spatial position of each patch in the path category dataset, identify adjacent land object patches connected by the boundary, and extract the numbers of the adjacent patches to obtain an adjacent patch index set; A land class change determination submodule: based on the adjacent patch index set, collect the land class numbers of adjacent patches at two remote sensing image time points, compare the number change situation, and summarize the number of patches with changes to obtain an adjacent land class change count result; A disturbance quantity calculation submodule: call the adjacent land class change count result, combine the number of directional connections between each patch boundary and adjacent patches at the current time point, calculate the correlation quantity of the number of adjacent patch land class changes and the number of connection directions, and obtain the land class disturbance factor corresponding to each patch.

7. The remote sensing driven land use monitoring system of claim 1, wherein: The land change monitoring module comprises: A path calling submodule: call the path type corresponding to each patch in the path category dataset, identify the directional adjustment path or the structure recognition path, obtain the main direction angle and the boundary contour similarity value of the patch in two remote sensing image periods according to the path type, and obtain a path type parameter set; A parameter comparison submodule: based on the path type parameter set, combine the land class disturbance factor of each patch, set a morphological parameter change judgment benchmark matched with the path type, calculate the parameter change amplitude of each patch in two periods, and generate a patch change comparison quantity; A change judgment submodule: compare the patch change comparison quantity with the set judgment benchmark, assign change state labels and stable state labels to the patches according to the comparison result, and obtain a land use change monitoring result.

8. The remote-sensing driven land use monitoring system of claim 7, wherein: For the boundary contour similarity value of the patch in two remote sensing image periods, the formula is: ； wherein, is the contour similarity of the target plaque at two time points, with a value range between 0 and 1, the closer to 1 means the closer of the contours, is the number of comparable boundary points in two time points, is the Euclidean distance of the corresponding boundary points between two time points, , are the contained areas of the target plaque boundary contour at two time points, respectively, is the maximum envelope area in the study region, is the boundary shape complexity weight, is the area change sensitivity weight.

9. The remote-sensing driven land use monitoring system of claim 1, wherein: It also includes a change type discrimination module: based on the land use change monitoring result, refine the land use change type identification to obtain a change classification result; The change classification result includes cultivated land change types, forest land change types, water body change types, and construction land change types.

10. The remote-sensing driven land use monitoring system of claim 9, wherein: The change type discrimination module comprises: A patch screening submodule: extract the patch numbers with change state labels in the land use change monitoring result, and determine the image positioning information in the current period remote sensing image to obtain a change patch index set; A feature extraction submodule: based on the change patch index set, obtain the principal component feature data and spectral reflection information of the corresponding patch in the current period remote sensing image, and collect the land class number of the patch in the last period to form a change patch feature combination data; A type determination submodule: according to the class difference between the current land class and the historical land class, the spectral value change direction, and the spatial structure change amplitude in the change patch feature combination data, the change patch is divided into multiple categories to obtain a change classification result.

Images