An illegal construction intelligent identification method and system

Abstract

The application relates to an illegal construction intelligent identification method and system, wherein the method performs an aerial photography task on a supervision area through a drone, generates a current aerial photography image, performs spatial alignment on a planning red line graph of the supervision area and the current aerial photography image based on pos data of the current aerial photography image, generates a full-range red line perception graph, respectively inputs historical aerial photography images and the current aerial photography image into a ground feature identification model for ground feature identification, obtains a suspected change graph according to obtained historical ground feature identification results and current ground feature identification results, extracts a construction feature of the current aerial photography image through a visual recognition algorithm to obtain a current construction distribution graph, superimposes the current construction distribution graph, the suspected change graph and the full-range red line perception graph, and regards an existing superimposed area as an illegal construction area. Therefore, the application realizes accurate, intelligent and objective identification of illegal construction, gets rid of artificial dependence, shortens a response cycle, and improves supervision efficiency of illegal construction.

Description

Technical Field

[0001] This invention relates to the field of urban management technology, and in particular to an intelligent identification method and system for illegal construction. Background Technology

[0002] In the process of urban management, the supervision of illegal construction mainly relies on manual ground patrols. However, due to its limited coverage, long response cycle, and strong subjectivity, it is difficult to adapt to large-scale and highly dynamic supervision scenarios.

[0003] In recent years, drone aerial photography technology has been gradually applied to regulatory scenarios. However, it mainly focuses on comparing simple pixels of images from different periods, and can only identify areas that have changed. The determination of whether the changed areas are illegal constructions still requires manual on-site verification. Although the coverage efficiency has been improved, there are still problems such as long response time and strong subjectivity, which affect the efficiency of monitoring illegal constructions. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: the present invention provides an intelligent identification method and system for illegal construction, which realizes accurate, intelligent and objective identification of illegal construction, eliminates reliance on manual labor, reduces subjectivity, shortens the response cycle and improves the supervision efficiency of illegal construction.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides an intelligent identification method for illegal construction, comprising: The drone performs aerial photography missions over the monitored area according to a preset cycle, generates current aerial images of the monitored area, acquires POS data of the current aerial images and a planning red line map of the monitored area, and spatially aligns the planning red line map with the current aerial images based on the POS data to generate an all-round red line perception map. Historical aerial images of the monitored area are acquired, and the historical aerial images and the current aerial images are respectively input into the ground feature recognition model for ground feature recognition to obtain historical ground feature recognition results and current ground feature recognition results. Based on the historical ground feature recognition results and the current ground feature recognition results, a suspected change map is obtained. Construction features are extracted from the current aerial image using a visual recognition algorithm. A current construction distribution map is obtained based on these features. The current construction distribution map, the suspected change map, and the all-round red line perception map are overlaid to determine if there is an overlaid area. If so, the overlaid area is identified as an illegal construction area.

[0006] The beneficial effects of this invention are as follows: Based on POS data, the planned red line map of the monitored area is spatially aligned with the current aerial imagery to generate a comprehensive red line perception map, solving the problems of strong subjectivity and long response cycles in traditional manual drawing comparison. By superimposing the current construction distribution map, suspected change map, and comprehensive red line perception map, illegal construction areas are identified. Compared to the traditional method of simply comparing pixels of images from different periods to identify changed areas, this invention forms a three-dimensional judgment logic that considers the authenticity of construction activities, the rationality of spatiotemporal changes, and the legality of the planned red line. It eliminates reliance on manual methods, achieving precise, intelligent, and objective identification of illegal construction areas, thus improving the accuracy of identification and the efficiency of monitoring illegal construction areas.

[0007] Optionally, the step of spatially aligning the planned redline map with the current aerial imagery based on the pos data to generate an all-around redline perception map includes: The SIFT algorithm is used to extract feature points from each frame of the current aerial image, and all feature points are matched to obtain the matched feature points. Based on the matched feature points and the pos data, aerial triangulation is performed on each frame of aerial photograph to obtain all frames of aerial photograph after aerial triangulation. The MVS algorithm is used to perform stereo calculations on all frames of aerial photographs after aerial triangulation to generate a 3D point cloud. The flight parameters of the UAV are obtained, and the current aerial image is differentially corrected based on the flight parameters and the three-dimensional point cloud to obtain the differentially corrected current aerial image. A two-dimensional orthophoto map of the monitored area is generated based on the differentially corrected current aerial image. By using coordinate transformation and affine transformation methods, the planned redline map and the two-dimensional orthophoto map are spatially aligned to generate an all-round redline perception map.

[0008] As described above, the SIFT algorithm extracts and matches feature points from each frame of aerial photograph. Compared to traditional corner detection algorithms, the SIFT algorithm is scale-invariant and rotation-invariant. Even if there are differences in shooting angle and size in the aerial photograph, it can still accurately extract and match feature points, ensuring the stability of feature point matching. Aerial triangulation is performed using the matched feature points and pos data to improve the spatial position accuracy of the aerial photograph. Based on flight parameters, differential correction is performed on the 3D point cloud obtained by the MVS algorithm to eliminate image distortion interference and improve the accuracy of the generated 2D orthophoto map of the monitored area. The dual transformation method of coordinate transformation and affine transformation achieves accurate matching between the planned redline map and the current aerial photograph, improving the accuracy of the obtained all-round redline perception map.

[0009] Optionally, the step of inputting the historical aerial images and the current aerial images into the ground feature recognition model for ground feature recognition, obtaining historical ground feature recognition results and current ground feature recognition results, and obtaining a suspected change map based on the historical ground feature recognition results and current ground feature recognition results includes: The historical aerial images and the current aerial images are divided into N corresponding historical tile images and N current tile images according to preset pixel sizes; The feature recognition model performs feature recognition on each historical tile image and each current tile image to obtain historical feature recognition results for each historical tile image and current feature recognition results for each current tile image. The historical feature recognition results and the current feature recognition results include vegetation and non-vegetation. Non-vegetation includes bare land, pits, and buildings. The historical feature identification results of each historical tile image are compared with the current feature identification results of the corresponding current tile image to obtain the current tile image that has changed. It is then determined whether the current tile image that has changed has changed from vegetation to non-vegetation. If so, a suspected change map is obtained based on the current tile image that has changed.

[0010] As described above, dividing historical and current aerial imagery into historical and current tile images for feature identification, compared to directly identifying all historical and current aerial imagery—that is, compared to global comparison—reduces computational burden and improves processing efficiency. Furthermore, tile-by-tile comparison accurately captures subtle changes, avoiding local omissions that can occur with global comparison. When current tile images show changes, a secondary classification from vegetation to non-vegetation can capture preparatory actions for illegal construction, improving the accuracy and comprehensiveness of illegal construction area identification.

[0011] Optionally, obtaining the suspected change map based on the changed current tile image includes: The current tile image that has changed and the corresponding historical tile image are converted to grayscale to obtain the current grayscale change map and the historical grayscale change map; Calculate the gray-level co-occurrence matrix of the current gray-level change map and the historical gray-level change map respectively to obtain the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix; Entropy, contrast, and homogeneity are extracted from the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix respectively to obtain the corresponding current entropy, current contrast, current homogeneity, historical entropy, historical contrast, and historical homogeneity. A first threshold is obtained based on the historical entropy value and a first preset parameter; a second threshold is obtained based on the historical contrast and a second preset parameter; and a third threshold is obtained based on the historical homogeneity and a third preset parameter. Determine whether the current entropy value is less than a first threshold, and simultaneously determine whether the current contrast is less than a second threshold, and simultaneously determine whether the current homogeneity is greater than a third threshold; If the current entropy value is less than the first threshold and the current contrast is less than the second threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current entropy value is less than the first threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current contrast is less than the second threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed.

[0012] As described above, by extracting entropy, comparison degree, and homogeneity from the gray-level co-occurrence matrix to perform multi-dimensional cross-validation of suspected change maps, vegetation changes not caused by construction can be effectively excluded, improving the accuracy of the obtained suspected change maps. Furthermore, all thresholds are based on dynamic changes in historical features, and the adaptive threshold method improves the adaptability of the scene, ensuring the authenticity and rationality of the obtained suspected change maps.

[0013] Optionally, the step of designating the superimposed area as an illegal construction area includes: Determine whether the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map. If the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map, the overlapping area is regarded as an illegal construction area, and the non-overlapping area is regarded as a suspected area. Obtain the list of legal construction projects corresponding to the monitored area and the current timestamp of the current aerial image. Match the suspected area and the current timestamp with the legal construction scope and corresponding validity period of all legal construction plans in the list of legal construction projects. If the match is not completely successful, the suspected area is regarded as an illegal construction area.

[0014] As described above, performing hierarchical judgment on the superimposed areas can avoid misjudgments caused by coarse-grained judgments and omissions caused by small-area boundary violations, thereby improving the accuracy of identifying illegal construction areas.

[0015] Secondly, the present invention provides an intelligent identification system for illegal construction, comprising: The data acquisition module is used for the UAV to perform aerial photography tasks on the monitored area according to a preset cycle, generate the current aerial image of the monitored area, acquire the pos data of the current aerial image and the planning red line map of the monitored area, and spatially align the planning red line map with the current aerial image based on the pos data to generate an all-round red line perception map. The feature recognition module is used to acquire historical aerial images of the monitored area, input the historical aerial images and the current aerial images into the feature recognition model for feature recognition, obtain historical feature recognition results and current feature recognition results, and obtain a suspected change map based on the historical feature recognition results and the current feature recognition results; The illegal construction area identification module is used to extract construction features from the current aerial image using a visual recognition algorithm, obtain a current construction distribution map based on the construction features, overlay the current construction distribution map, the suspected change map, and the all-round red line perception map, and determine whether there is an overlay area. If there is, the overlay area is identified as an illegal construction area.

[0016] The beneficial effects of this invention are as follows: Based on POS data, the planned red line map of the monitored area is spatially aligned with the current aerial imagery to generate a comprehensive red line perception map, solving the problems of strong subjectivity and long response cycles in traditional manual drawing comparison. By superimposing the current construction distribution map, suspected change map, and comprehensive red line perception map, illegal construction areas are identified. Compared to the traditional method of simply comparing pixels of images from different periods to identify changed areas, this invention forms a three-dimensional judgment logic that considers the authenticity of construction activities, the rationality of spatiotemporal changes, and the legality of the planned red line. It eliminates reliance on manual methods, achieving precise, intelligent, and objective identification of illegal construction areas, thus improving the accuracy of identification and the efficiency of monitoring illegal construction areas.

[0017] Optionally, the acquisition module includes: The all-round redline perception map generation module is used to extract feature points of each frame of the current aerial image using the SIFT algorithm, and match all feature points to obtain the matched feature points. Based on the matched feature points and the pos data, aerial triangulation is performed on each frame of aerial photograph to obtain all frames of aerial photograph after aerial triangulation. The MVS algorithm is used to perform stereo calculations on all frames of aerial photographs after aerial triangulation to generate a 3D point cloud. The flight parameters of the UAV are obtained, and the current aerial image is differentially corrected based on the flight parameters and the three-dimensional point cloud to obtain the differentially corrected current aerial image. A two-dimensional orthophoto map of the monitored area is generated based on the differentially corrected current aerial image. By using coordinate transformation and affine transformation methods, the planned redline map and the two-dimensional orthophoto map are spatially aligned to generate an all-round redline perception map.

[0018] As described above, the SIFT algorithm extracts and matches feature points from each frame of aerial photograph. Compared to traditional corner detection algorithms, the SIFT algorithm is scale-invariant and rotation-invariant. Even if there are differences in shooting angle and size in the aerial photograph, it can still accurately extract and match feature points, ensuring the stability of feature point matching. Aerial triangulation is performed using the matched feature points and pos data to improve the spatial position accuracy of the aerial photograph. Based on flight parameters, differential correction is performed on the 3D point cloud obtained by the MVS algorithm to eliminate image distortion interference and improve the accuracy of the generated 2D orthophoto map of the monitored area. The dual transformation method of coordinate transformation and affine transformation achieves accurate matching between the planned redline map and the current aerial photograph, improving the accuracy of the obtained all-round redline perception map.

[0019] Optionally, the feature identification module includes: The change perception module is used to divide the historical aerial images and the current aerial images into N corresponding historical tile images and N current tile images according to a preset pixel size; The feature recognition model performs feature recognition on each historical tile image and each current tile image to obtain historical feature recognition results for each historical tile image and current feature recognition results for each current tile image. The historical feature recognition results and the current feature recognition results include vegetation and non-vegetation. Non-vegetation includes bare land, pits, and buildings. The historical feature identification results of each historical tile image are compared with the current feature identification results of the corresponding current tile image to obtain the current tile image that has changed. It is then determined whether the current tile image that has changed has changed from vegetation to non-vegetation. If so, a suspected change map is obtained based on the current tile image that has changed.

[0020] As described above, dividing historical and current aerial imagery into historical and current tile images for feature identification, compared to directly identifying all historical and current aerial imagery—that is, compared to global comparison—reduces computational burden and improves processing efficiency. Furthermore, tile-by-tile comparison accurately captures subtle changes, avoiding local omissions that can occur with global comparison. When current tile images show changes, a secondary classification from vegetation to non-vegetation can capture preparatory actions for illegal construction, improving the accuracy and comprehensiveness of illegal construction area identification.

[0021] Optionally, the change sensing module specifically comprises: The current tile image that has changed and the corresponding historical tile image are converted to grayscale to obtain the current grayscale change map and the historical grayscale change map; Calculate the gray-level co-occurrence matrix of the current gray-level change map and the historical gray-level change map respectively to obtain the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix; Entropy, contrast, and homogeneity are extracted from the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix respectively to obtain the corresponding current entropy, current contrast, current homogeneity, historical entropy, historical contrast, and historical homogeneity. A first threshold is obtained based on the historical entropy value and a first preset parameter; a second threshold is obtained based on the historical contrast and a second preset parameter; and a third threshold is obtained based on the historical homogeneity and a third preset parameter. Determine whether the current entropy value is less than a first threshold, and simultaneously determine whether the current contrast is less than a second threshold, and simultaneously determine whether the current homogeneity is greater than a third threshold; If the current entropy value is less than the first threshold and the current contrast is less than the second threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current entropy value is less than the first threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current contrast is less than the second threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed.

[0022] As described above, by extracting entropy, comparison degree, and homogeneity from the gray-level co-occurrence matrix to perform multi-dimensional cross-validation of suspected change maps, vegetation changes not caused by construction can be effectively excluded, improving the accuracy of the obtained suspected change maps. Furthermore, all thresholds are based on dynamic changes in historical features, and the adaptive threshold method improves the adaptability of the scene, ensuring the authenticity and rationality of the obtained suspected change maps.

[0023] Optionally, the illegal construction area identification module specifically comprises: Determine whether the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map. If the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map, the overlapping area is regarded as an illegal construction area, and the non-overlapping area is regarded as a suspected area. Obtain the list of legal construction projects corresponding to the monitored area and the current timestamp of the current aerial image. Match the suspected area and the current timestamp with the legal construction scope and corresponding validity period of all legal construction plans in the list of legal construction projects. If the match is not completely successful, the suspected area is regarded as an illegal construction area.

[0024] As described above, performing hierarchical judgment on the superimposed areas can avoid misjudgments caused by coarse-grained judgments and omissions caused by small-area boundary violations, thereby improving the accuracy and comprehensiveness of illegal construction area identification. Attached Figure Description

[0025] Figure 1 A flowchart illustrating an intelligent method for identifying illegal construction provided in this embodiment; Figure 2 This is a schematic diagram of the overall process of an intelligent identification method for illegal construction provided in this embodiment; Figure 3 This is a schematic diagram of the structure of an intelligent identification system for illegal construction provided in this embodiment.

[0026] [Explanation of Labels in the Attached Image] 1. An intelligent identification system for illegal construction; 2. Data Acquisition Module; 21. All-Around Red Line Perception Map Generation Module; 3. Ground feature recognition module; 31. Change sensing module; 4. Illegal construction area identification module. Detailed Implementation

[0027] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0028] Example 1 Please refer to Figures 1 to 2 This invention provides an intelligent identification method for illegal construction, comprising the following steps: S1. The drone performs aerial photography tasks on the monitored area according to a preset cycle, generates the current aerial image of the monitored area, obtains the pos data of the current aerial image and the planning red line map of the monitored area, and spatially aligns the planning red line map with the current aerial image based on the pos data to generate an all-round red line perception map. In this embodiment, as Figure 2As shown, the drone performs aerial photography missions over the monitored area according to a preset cycle, generating current aerial images of the monitored area. The preset cycle is one week, but it can be adjusted according to actual conditions. The current aerial images carry POS data. Each monitored area has its own planning red line map, which contains specific coordinate information. Based on the POS data, the current aerial images are spatially aligned with the planning red line map to generate a comprehensive red line perception map.

[0029] At this point, step S1, which involves spatially aligning the planned redline map with the current aerial image based on the pos data to generate an all-around redline perception map, includes: S11. Extract feature points of each frame of the current aerial image using the SIFT algorithm, and match all feature points to obtain the matched feature points. S12. Perform aerial triangulation on each frame of aerial photograph based on the matched feature points and the pos data to obtain all frames of aerial photograph after aerial triangulation. S13. Perform stereo calculations on all frames of aerial photographs after aerial triangulation using the MVS algorithm to generate a three-dimensional point cloud. S14. Obtain the flight parameters of the UAV, perform differential correction on the current aerial image based on the flight parameters and the three-dimensional point cloud to obtain the differentially corrected current aerial image, and generate a two-dimensional orthophoto map of the monitored area based on the differentially corrected current aerial image. S15. Using coordinate transformation and affine transformation, the planned redline map and the two-dimensional orthophoto map are spatially aligned to generate an all-round redline perception map.

[0030] In this embodiment, as Figure 2As shown, the SIFT (Scale-invariant feature transform) algorithm is used to extract feature points from each frame of the current aerial imagery. All feature points are then matched, specifically those with the same name. Based on these matched feature points and position data, aerial triangulation is performed on each frame, resulting in a triangulated aerial network that calculates the spatial position of each frame. Finally, the MVS algorithm is used to perform stereo calculations on all the triangulated frames, generating a 3D point cloud where each point contains 3D coordinates. The process involves acquiring the drone's flight parameters, including but not limited to altitude, angle, resolution, forward overlap, and lateral overlap. Based on these parameters and the 3D point cloud, differential correction is performed on the current aerial imagery. This involves gridding the 3D point cloud, preserving the highest point elevation of each grid cell, and generating a Digital Representation Model (DSM). Based on the elevation information provided by the DSM and the flight parameters, differential correction is applied to the current aerial imagery to eliminate image distortion caused by terrain undulations and camera tilt. Then, each frame of the corrected aerial imagery is stitched together according to geographic coordinates to generate a 2D orthophoto map of the monitored area. The coordinates of the planned redline map are transformed to match those of the 2D orthophoto map using a coordinate transformation method, and then spatial alignment is achieved using an affine transformation method to generate an all-around redline perception map.

[0031] S2. Obtain historical aerial images of the monitored area, input the historical aerial images and the current aerial images into the ground feature recognition model for ground feature recognition, obtain historical ground feature recognition results and current ground feature recognition results, and obtain a suspected change map based on the historical ground feature recognition results and current ground feature recognition results; In this embodiment, as Figure 2 As shown, historical aerial images of the monitored area are obtained. The historical aerial images and the current aerial images are then input into the ground feature recognition model for ground feature recognition. At this time, the ground feature recognition model is a semantic segmentation model pre-built based on a deep learning network architecture. Based on the obtained historical ground feature recognition results and the obtained current ground feature recognition results, a suspected change map is obtained.

[0032] At this point, step S2, which involves inputting the historical aerial imagery and the current aerial imagery into the feature recognition model for feature recognition to obtain historical feature recognition results and current feature recognition results, and then obtaining a suspected change map based on the historical feature recognition results and the current feature recognition results, includes: S21. Divide the historical aerial images and the current aerial images into N corresponding historical tile images and N current tile images according to the preset pixel size; S22. The feature recognition model performs feature recognition on each historical tile image and each current tile image to obtain the historical feature recognition result for each historical tile image and the current feature recognition result for each current tile image. The historical feature recognition result and the current feature recognition result include vegetation and non-vegetation. The non-vegetation includes bare land, pits and buildings. S23. Compare the historical feature identification results of each historical tile image with the current feature identification results of the corresponding current tile image to obtain the current tile image that has changed. Determine whether the current tile image that has changed has changed from vegetation to non-vegetation. If so, obtain a suspected change map based on the current tile image that has changed.

[0033] In this embodiment, as Figure 2 As shown, historical and current aerial images are divided into N historical tile images and N current tile images according to a preset pixel size of 512×512, which can be adjusted according to actual conditions. The feature recognition model performs feature recognition on each historical tile image and each current tile image, obtaining historical feature recognition results for each historical tile image and current feature recognition results for each current tile image. Feature recognition results include vegetation and non-vegetation, with non-vegetation including bare land, pits, and buildings. The historical feature recognition results of each historical tile image are compared with the corresponding current feature recognition results of the current tile image to obtain the current tile image that has changed, and the specific changes can be determined. If the changed current tile image changes from vegetation to non-vegetation, a suspected change map is obtained. In this embodiment, suspected change maps are classified according to the specific changes to facilitate further identification of the construction stage of illegal construction areas when determining illegal construction areas, allowing for targeted processing. The specific classification rules for suspected change maps are as follows: 1. When the change is from vegetation to bare land, it is marked as: Preliminary suspected change map, and the corresponding construction stage is: Construction preparation stage; 2. When the change is from vegetation to a pit, it is marked as: suspected change map during the excavation period, and the corresponding construction stage is: foundation pit excavation period; 3. When the change is from vegetation to buildings, it is marked as: suspected change map during the construction period, and the corresponding construction stage is: main construction period.

[0034] At this point, step S23, which involves obtaining a suspected change map based on the changed current tile image, includes: S231. Perform grayscale processing on the current tile image that has changed and the corresponding historical tile image to obtain the current grayscale change map and the historical grayscale change map; S232. Calculate the gray-level co-occurrence matrix of the current gray-level change map and the historical gray-level change map respectively to obtain the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix; S233. Extract the entropy, contrast and homogeneity of the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix respectively to obtain the corresponding current entropy, current contrast, current homogeneity, historical entropy, historical contrast and historical homogeneity. S234. A first threshold is obtained based on the historical entropy value and a first preset parameter; a second threshold is obtained based on the historical contrast and a second preset parameter; and a third threshold is obtained based on the historical homogeneity and a third preset parameter. S235. Determine whether the current entropy value is less than the first threshold, and at the same time determine whether the current contrast is less than the second threshold, and at the same time determine whether the current homogeneity is greater than the third threshold. S236. If the current entropy value is less than the first threshold and the current contrast is less than the second threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or S237. If the current entropy value is less than the first threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or S238. If the current contrast is less than the second threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed.

[0035] In this embodiment, to avoid misclassifying non-construction change images as suspected change images (e.g., mistaking natural changes in vegetation for construction work), multi-dimensional cross-validation is performed. Before grayscale processing of the changed current tile image and its corresponding historical tile image, an elevation difference calculation is performed on both. Specifically, the elevation value of the changed current tile image is subtracted from the corresponding elevation value of the historical tile image to obtain an elevation difference map. The average value of pixels within the elevation difference map and the percentage of negatively changing pixels are calculated. If the calculated average value is less than an average threshold and the calculated percentage of negatively changing pixels is greater than a pixel threshold, the changed current tile image is retained; otherwise, it is filtered out. The average threshold is 0.3, and the pixel threshold is 60%. These can be adjusted according to actual conditions.

[0036] By converting the changed current tile image and its corresponding historical tile image to grayscale, we obtain the current grayscale change map and the historical grayscale change map. We then calculate the grayscale co-occurrence matrix (HCMM) for both, resulting in the current HCMM and the historical HCMM. Finally, we extract the entropy, contrast, and homogeneity of both the current and historical HCMMs, yielding the corresponding current entropy, current contrast, current homogeneity, historical entropy, historical contrast, and historical homogeneity. During cross-validation, the threshold is based on historical features, specifically, the historical entropy, historical contrast, and historical homogeneity dynamically change. A first threshold is obtained based on the historical entropy and a first preset parameter (1.5). A second threshold is obtained based on the historical contrast and a second preset parameter (0.4). A third threshold is obtained based on the historical homogeneity and a third preset parameter (0.2). The first, second, and third preset parameters can all be adjusted according to the actual situation. The algorithm determines whether the current entropy value is less than the first threshold, whether the current contrast is less than the second threshold, and whether the current homogeneity is greater than the third threshold. If any two of these conditions are met, a suspected change image can be obtained from the changed current tile image.

[0037] S3. Extract construction features from the current aerial image using a visual recognition algorithm, obtain a current construction distribution map based on the construction features, overlay the current construction distribution map, the suspected change map, and the all-round red line perception map, and determine whether there is an overlay area. If there is, the overlay area is taken as an illegal construction area.

[0038] In this embodiment, as Figure 2 As shown, construction features are extracted from the current aerial image using a visual recognition algorithm. These features include construction equipment, construction signs, and construction traces. A current construction distribution map is obtained based on these features. The current construction distribution map, the suspected change map, and the all-round red line perception map are then overlaid, i.e., triple overlay. The system then determines whether there is an overlay area. If so, the overlay area is identified as an illegal construction area.

[0039] At this point, step S3, which defines the superimposed area as an illegal construction area, includes: S31. Determine whether the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map. If the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map, the overlapping area is regarded as an illegal construction area, and the non-overlapping area is regarded as a suspected area. S32. Obtain the list of legal construction projects corresponding to the monitored area and the current timestamp of the current aerial image. Simultaneously match the suspected area and the current timestamp with the legal construction scope and corresponding validity period of all legal construction plans in the list of legal construction projects. If the match is not completely successful, the suspected area is regarded as an illegal construction area.

[0040] In this embodiment, as Figure 2 As shown, the overlapping areas are layered for judgment. The all-round red line perception map is divided into prohibited construction areas and permitted construction areas. If the overlapping area overlaps with the prohibited area, the overlapping area is an illegal construction area. The non-overlapping area is considered a suspected area for further judgment. A list of legal construction projects corresponding to the regulated area is obtained. The list of legal construction projects includes each legal construction plan, as well as the validity period and legal construction scope of each legal construction plan. The suspected area and the current timestamp of the current aerial image are matched with the legal construction specifications and corresponding validity period of each legal construction plan. If the match is successful, it means that the suspected area is actually within the permitted construction scope and is a legally reported construction plan. Otherwise, if the match is not successful, the suspected area is an illegal construction area. For example: If a suspected area matches the legal construction specifications of a legal construction plan A, but the current timestamp does not match the validity period of legal construction plan A, then the suspected area may be an overdue construction area, which is also a type of illegal construction; that is, the suspected area is an illegal construction area. If the current timestamp does not match the validity period of legal construction plan B, but the suspected area matches the legal construction specifications of legal construction plan B, then it is an over-scope construction area, which is also a type of illegal construction; that is, the suspected area is an illegal construction area. If a suspected area does not match the legal construction specifications of legal construction plan C, and the current timestamp does not match the validity period of legal construction plan C, then the suspected area may be an unapproved construction area, which is also a type of illegal construction; that is, the suspected area is an illegal construction area.

[0041] In this embodiment, the risk level of the obtained illegal construction area can be calculated. This is achieved by calculating the area of ​​the illegal construction area, the rate of change of the superimposed area compared to the previous period, and the rate of change of the suspected change map compared to the previous period. These calculated factors are then input into the risk level formula to obtain the risk level of the illegal construction area. Differentiated warnings are issued for different levels of risk. The risk level formula is as follows: Suspected rate of change; Where R represents the degree of risk, Indicates the first weight. Indicates the second weight. This indicates the third weight.

[0042] Example 2 Please refer to Figure 3 The present invention provides an intelligent identification system 1 for illegal construction, comprising: a data acquisition module 2, an all-round red line perception map generation module 21, a ground feature identification module 3, a change perception module 31, and an illegal construction area identification module 4.

[0043] Among them, the acquisition module 2 is used for the UAV to perform aerial photography tasks on the monitored area according to a preset cycle, generate the current aerial image of the monitored area, acquire the pos data of the current aerial image and the planning red line map of the monitored area, and spatially align the planning red line map with the current aerial image based on the pos data to generate an all-round red line perception map. The feature recognition module 3 is used to acquire historical aerial images of the monitored area, input the historical aerial images and the current aerial images into the feature recognition model for feature recognition, obtain historical feature recognition results and current feature recognition results, and obtain a suspected change map based on the historical feature recognition results and current feature recognition results; The illegal construction area identification module 4 is used to extract construction features from the current aerial image through a visual recognition algorithm, obtain a current construction distribution map based on the construction features, overlay the current construction distribution map, the suspected change map, and the all-round red line perception map, and determine whether there is an overlay area. If there is, the overlay area is identified as an illegal construction area.

[0044] Specifically, the acquisition module 2 includes: The all-round redline perception map generation module 21 is used to extract feature points of each frame of the current aerial image through the SIFT algorithm, and match all feature points to obtain the matched feature points. Based on the matched feature points and the pos data, aerial triangulation is performed on each frame of aerial photograph to obtain all frames of aerial photograph after aerial triangulation. The MVS algorithm is used to perform stereo calculations on all frames of aerial photographs after aerial triangulation to generate a 3D point cloud. The flight parameters of the UAV are obtained, and the current aerial image is differentially corrected based on the flight parameters and the three-dimensional point cloud to obtain the differentially corrected current aerial image. A two-dimensional orthophoto map of the monitored area is generated based on the differentially corrected current aerial image. By using coordinate transformation and affine transformation methods, the planned redline map and the two-dimensional orthophoto map are spatially aligned to generate an all-round redline perception map.

[0045] Specifically, the feature identification module 3 includes: The change perception module 31 is used to divide the historical aerial images and the current aerial images into N corresponding historical tile images and N current tile images according to a preset pixel size; The feature recognition model performs feature recognition on each historical tile image and each current tile image to obtain historical feature recognition results for each historical tile image and current feature recognition results for each current tile image. The historical feature recognition results and the current feature recognition results include vegetation and non-vegetation. Non-vegetation includes bare land, pits, and buildings. The historical feature identification results of each historical tile image are compared with the current feature identification results of the corresponding current tile image to obtain the current tile image that has changed. It is then determined whether the current tile image that has changed has changed from vegetation to non-vegetation. If so, a suspected change map is obtained based on the current tile image that has changed.

[0046] Specifically, the change sensing module 31 is as follows: The current tile image that has changed and the corresponding historical tile image are converted to grayscale to obtain the current grayscale change map and the historical grayscale change map; Calculate the gray-level co-occurrence matrix of the current gray-level change map and the historical gray-level change map respectively to obtain the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix; Entropy, contrast, and homogeneity are extracted from the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix respectively to obtain the corresponding current entropy, current contrast, current homogeneity, historical entropy, historical contrast, and historical homogeneity. A first threshold is obtained based on the historical entropy value and a first preset parameter; a second threshold is obtained based on the historical contrast and a second preset parameter; and a third threshold is obtained based on the historical homogeneity and a third preset parameter. Determine whether the current entropy value is less than a first threshold, and simultaneously determine whether the current contrast is less than a second threshold, and simultaneously determine whether the current homogeneity is greater than a third threshold; If the current entropy value is less than the first threshold and the current contrast is less than the second threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current entropy value is less than the first threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current contrast is less than the second threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed.

[0047] Specifically, the illegal construction area identification module 4 is as follows: Determine whether the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map. If the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map, the overlapping area is regarded as an illegal construction area, and the non-overlapping area is regarded as a suspected area. Obtain the list of legal construction projects corresponding to the monitored area and the current timestamp of the current aerial image. Match the suspected area and the current timestamp with the legal construction scope and corresponding validity period of all legal construction plans in the list of legal construction projects. If the match is not completely successful, the suspected area is regarded as an illegal construction area.

[0048] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0049] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0050] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0051] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0052] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0053] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0054] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A method for intelligent identification of illegal construction, characterized in that, include: The drone performs aerial photography missions over the monitored area according to a preset cycle, generates current aerial images of the monitored area, acquires POS data of the current aerial images and a planning red line map of the monitored area, and spatially aligns the planning red line map with the current aerial images based on the POS data to generate an all-round red line perception map. Historical aerial images of the monitored area are acquired, and the historical aerial images and the current aerial images are respectively input into the ground feature recognition model for ground feature recognition to obtain historical ground feature recognition results and current ground feature recognition results. Based on the historical ground feature recognition results and the current ground feature recognition results, a suspected change map is obtained. Construction features are extracted from the current aerial image using a visual recognition algorithm. A current construction distribution map is obtained based on these features. The current construction distribution map, the suspected change map, and the all-around redline perception map are overlaid. It is then determined whether an overlaid area exists. If so, the overlaid area is designated as an illegal construction area, including: Determine whether the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map. If the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map, the overlapping area is regarded as an illegal construction area, and the non-overlapping area is regarded as a suspected area. Obtain the list of legal construction projects corresponding to the monitored area and the current timestamp of the current aerial image. Match the suspected area and the current timestamp with the legal construction scope and corresponding validity period of all legal construction plans in the list of legal construction projects. If the match is not completely successful, the suspected area is regarded as an illegal construction area.

2. The intelligent identification method for illegal construction as described in claim 1, characterized in that, The step of spatially aligning the planned redline map with the current aerial imagery based on the pos data to generate a comprehensive redline perception map includes: The SIFT algorithm is used to extract feature points from each frame of the current aerial image, and all feature points are matched to obtain the matched feature points. Based on the matched feature points and the pos data, aerial triangulation is performed on each frame of aerial photograph to obtain all frames of aerial photograph after aerial triangulation. The MVS algorithm is used to perform stereo calculations on all frames of aerial photographs after aerial triangulation to generate a 3D point cloud. The flight parameters of the UAV are obtained, and the current aerial image is differentially corrected based on the flight parameters and the three-dimensional point cloud to obtain the differentially corrected current aerial image. A two-dimensional orthophoto map of the monitored area is generated based on the differentially corrected current aerial image. By using coordinate transformation and affine transformation, the planned redline map and the two-dimensional orthophoto map are spatially aligned to generate an all-round redline perception map.

3. The intelligent identification method for illegal construction as described in claim 1, characterized in that, The step of inputting the historical aerial images and the current aerial images into the ground feature recognition model for ground feature recognition, obtaining historical ground feature recognition results and current ground feature recognition results, and obtaining a suspected change map based on the historical ground feature recognition results and current ground feature recognition results includes: The historical aerial images and the current aerial images are divided into N corresponding historical tile images and N current tile images according to preset pixel sizes; The feature recognition model performs feature recognition on each historical tile image and each current tile image to obtain historical feature recognition results for each historical tile image and current feature recognition results for each current tile image. The historical feature recognition results and the current feature recognition results include vegetation and non-vegetation. Non-vegetation includes bare land, pits, and buildings. The historical feature identification results of each historical tile image are compared with the current feature identification results of the corresponding current tile image to obtain the current tile image that has changed. It is then determined whether the current tile image that has changed has changed from vegetation to non-vegetation. If so, a suspected change map is obtained based on the current tile image that has changed.

4. The intelligent identification method for illegal construction as described in claim 3, characterized in that, The process of obtaining a suspected change map based on the changed current tile image includes: The current tile image that has changed and the corresponding historical tile image are converted to grayscale to obtain the current grayscale change map and the historical grayscale change map; Calculate the gray-level co-occurrence matrix of the current gray-level change map and the historical gray-level change map respectively to obtain the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix; Entropy, contrast, and homogeneity are extracted from the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix respectively to obtain the corresponding current entropy, current contrast, current homogeneity, historical entropy, historical contrast, and historical homogeneity. A first threshold is obtained based on the historical entropy value and a first preset parameter; a second threshold is obtained based on the historical contrast and a second preset parameter; and a third threshold is obtained based on the historical homogeneity and a third preset parameter. Determine whether the current entropy value is less than a first threshold, and simultaneously determine whether the current contrast is less than a second threshold, and simultaneously determine whether the current homogeneity is greater than a third threshold; If the current entropy value is less than the first threshold and the current contrast is less than the second threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current entropy value is less than the first threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current contrast is less than the second threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed.

5. An intelligent identification system for illegal construction, characterized in that, include: The data acquisition module is used for the UAV to perform aerial photography tasks on the monitored area according to a preset cycle, generate the current aerial image of the monitored area, acquire the pos data of the current aerial image and the planning red line map of the monitored area, and spatially align the planning red line map with the current aerial image based on the pos data to generate an all-round red line perception map. The feature recognition module is used to acquire historical aerial images of the monitored area, input the historical aerial images and the current aerial images into the feature recognition model for feature recognition, obtain historical feature recognition results and current feature recognition results, and obtain a suspected change map based on the historical feature recognition results and the current feature recognition results; The illegal construction area identification module is used to extract construction features from the current aerial image using a visual recognition algorithm, obtain a current construction distribution map based on the construction features, overlay the current construction distribution map, the suspected change map, and the all-round red line perception map, and determine whether there is an overlay area. If there is, the overlay area is identified as an illegal construction area. The illegal construction area identification module is specifically as follows: Determine whether the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map. If the superimposed area overlaps with the prohibited construction area of ​​the all-round red line perception map, the overlapping area is regarded as an illegal construction area, and the non-overlapping area is regarded as a suspected area. Obtain the list of legal construction projects corresponding to the monitored area and the current timestamp of the current aerial image. Match the suspected area and the current timestamp with the legal construction scope and corresponding validity period of all legal construction plans in the list of legal construction projects. If the match is not completely successful, the suspected area is regarded as an illegal construction area.

6. The intelligent identification system for illegal construction as described in claim 5, characterized in that, The acquisition module includes: The all-round redline perception map generation module is used to extract feature points of each frame of the current aerial image using the SIFT algorithm, and match all feature points to obtain the matched feature points. Based on the matched feature points and the pos data, aerial triangulation is performed on each frame of aerial photograph to obtain all frames of aerial photograph after aerial triangulation. The MVS algorithm is used to perform stereo calculations on all frames of aerial photographs after aerial triangulation to generate a 3D point cloud. The flight parameters of the UAV are obtained, and the current aerial image is differentially corrected based on the flight parameters and the three-dimensional point cloud to obtain the differentially corrected current aerial image. A two-dimensional orthophoto map of the monitored area is generated based on the differentially corrected current aerial image. By using coordinate transformation and affine transformation, the planned redline map and the two-dimensional orthophoto map are spatially aligned to generate an all-round redline perception map.

7. The intelligent identification system for illegal construction as described in claim 5, characterized in that, The feature identification module includes: The change perception module is used to divide the historical aerial images and the current aerial images into N corresponding historical tile images and N current tile images according to a preset pixel size; The feature recognition model performs feature recognition on each historical tile image and each current tile image to obtain historical feature recognition results for each historical tile image and current feature recognition results for each current tile image. The historical feature recognition results and the current feature recognition results include vegetation and non-vegetation. Non-vegetation includes bare land, pits, and buildings. The historical feature identification results of each historical tile image are compared with the current feature identification results of the corresponding current tile image to obtain the current tile image that has changed. It is then determined whether the current tile image that has changed has changed from vegetation to non-vegetation. If so, a suspected change map is obtained based on the current tile image that has changed.

8. The intelligent identification system for illegal construction as described in claim 7, characterized in that, The change sensing module is specifically: The current tile image that has changed and the corresponding historical tile image are converted to grayscale to obtain the current grayscale change map and the historical grayscale change map; Calculate the gray-level co-occurrence matrix of the current gray-level change map and the historical gray-level change map respectively to obtain the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix; Entropy, contrast, and homogeneity are extracted from the current gray-level co-occurrence matrix and the historical gray-level co-occurrence matrix respectively to obtain the corresponding current entropy, current contrast, current homogeneity, historical entropy, historical contrast, and historical homogeneity. A first threshold is obtained based on the historical entropy value and a first preset parameter; a second threshold is obtained based on the historical contrast and a second preset parameter; and a third threshold is obtained based on the historical homogeneity and a third preset parameter. Determine whether the current entropy value is less than a first threshold, and simultaneously determine whether the current contrast is less than a second threshold, and simultaneously determine whether the current homogeneity is greater than a third threshold; If the current entropy value is less than the first threshold and the current contrast is less than the second threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current entropy value is less than the first threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed. and / or If the current contrast is less than the second threshold and the current homogeneity is greater than the third threshold, then a suspected change map is obtained based on the current tile image that has changed.

Images