A method for identifying anthropogenic disturbances of land

Abstract

The application discloses a kind of identification methods for human disturbance land, it is related to remote sensing image field, the method is by in edge computing terminal to the unified geometric orthographic correction of remote sensing image sequence, radiation normalization, cloud mask processing and format conversion, construct standardization image set, effectively improve the time series comparability and spectral consistency between remote sensing image, solve multi-temporal image illumination difference, geographical deviation and cloud interference and other influencing factors, provide high-quality input data basis for subsequent disturbance feature extraction;Meanwhile, the application combines air port identification algorithm, mechanical activity path detection and construction boundary expansion identification technology, comprehensively extracts ventilation corridor area, dynamic expansion area and mechanical disturbance area, constructs sensitive disturbance sampling point space layout scheme for disturbance potential, significantly improves the perception coverage ability and detection efficiency to key area, reduces redundant computing resource consumption, improves system real-time processing capability and early identification precision.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image technology, specifically to a method for identifying land use disturbed by human activity. Background Technology

[0002] With the rapid development of remote sensing technology and the acquisition of large amounts of high-resolution, multi-temporal remote sensing images, land surface change detection based on remote sensing images has been widely applied in environmental monitoring, ecological protection, and land use supervision. Especially under the promotion of relevant authorities such as the Ministry of Natural Resources and the Ministry of Ecology and Environment, remote sensing has become a key application area for land surface disturbance monitoring in arid and semi-arid regions. In recent years, due to frequent human activities such as wind power development and highway construction, land surface disturbance problems have become increasingly prominent, urgently requiring high-precision and high-timeliness identification methods. Among these application needs, the identification of "dust-type disturbances caused by construction and transportation within human-caused areas" is the most urgent and challenging, because such disturbances usually manifest as short-term, intermittent, unstructured disturbances, which cannot be directly identified based on surface geometric changes.

[0003] Current remote sensing methods for identifying human-caused land disturbance in target areas mostly employ direct methods such as spectral change detection, surface differential analysis, or deep learning semantic segmentation to extract the disturbed area. However, due to the high reflectivity of the terrain, the susceptibility of the disturbed area to wind and sand cover, and the short-term, sudden nature of the disturbance, traditional direct detection methods based on surface changes have several problems: First, the reflectivity difference between the disturbed area and the background features is insufficient, making it difficult to form a significant differential response; second, some disturbances are subsequently covered by wind and sand, resulting in no significant geometric or textural changes in the remote sensing image; third, the disturbance behavior is not continuous, leading to a lack of significant consistency in the disturbance pattern across image time series, resulting in numerous missed and false identifications, and reducing the monitoring system's ability to detect and respond to key construction activities.

[0004] The difficulties in identifying disturbances using existing methods stem primarily from the unique physical environment and disturbance behavior characteristics of the affected areas. On one hand, the surface is often composed of exposed sandstone and conglomerate layers, with only slight spectral differences before and after disturbance, masking the change signals relied upon by traditional image recognition methods. On the other hand, construction or transportation activities are often short-lived and sudden; disturbing materials such as dust are easily dispersed or settled by wind within a few hours, making it impossible to capture the complete disturbance process at the time of image acquisition. Furthermore, strong winds and sandstorms often accompany disturbances, rapidly masking the direct results and making it impossible to detect disturbance boundaries or significant changes in remote sensing images. Ultimately, this identification failure leads to several problems: such as the failure to promptly mark sensitive construction areas, delayed environmental monitoring responses, and distorted environmental risk assessments, thereby affecting the implementation of regulatory policies and the accuracy of dynamic land use monitoring. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for identifying land use disturbed by human activities, which solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying land use disturbed by human activity, comprising the following steps:

[0007] S1. Collect remote sensing image sequences of the target area at multiple consecutive time points, and transmit the remote sensing image sequences to the edge computing terminal. Perform image preprocessing on the remote sensing image sequences in the edge computing terminal to obtain a standardized image set.

[0008] S2. Extract perturbation features from each image in the standardized image set to obtain an environmental perturbation dataset, and preprocess the environmental perturbation data to obtain a standardized environmental perturbation dataset.

[0009] S3. Calculate and output the perturbation confidence score index Dti based on the standardized environmental perturbation dataset, and conduct a preliminary comparative evaluation based on the perturbation confidence score index Dt, and classify the perturbation level.

[0010] S4. Based on the preliminary comparative evaluation results, trigger the time spectrum inversion process, extract N periods of remote sensing images of sensitive disturbance sampling points at level 2, obtain the image time series set, perform feature extraction to obtain the disturbance amplitude f, and then calculate and output the disturbance frequency coherence factor AdfC.

[0011] S5. The perturbation frequency coherence factor AdfC and the perturbation confidence score index Dti are calculated together to output the comprehensive perturbation index Dci. The perturbation risk interval threshold is preset and the comprehensive perturbation index Dci is compared and evaluated a second time. The corresponding strategy is executed based on the results of the second comparison and evaluation.

[0012] Preferably, S1 includes S11 and S12;

[0013] S11. By setting up API application interfaces, connect to the remote sensing satellite data service interface and the ERA5 meteorological analysis data platform respectively, and set the acquisition conditions for the API application interfaces to acquire remote sensing images of the target area in real time.

[0014] The acquisition conditions include acquiring multi-temporal image sequences over the past 60 days, with the images being captured at 12:00 ± 1 hour each day and cloud cover less than 10%, and the image bands covering the visible and near-infrared bands.

[0015] The wind corridor effect is analyzed in remote sensing images using a wind vent recognition algorithm to extract ventilation corridor areas;

[0016] A remote sensing image automatic rut identification algorithm is used to locate areas with frequent mechanical activity.

[0017] Edge change detection is performed by comparing remote sensing images from each period to identify the dynamic expansion area of ​​the construction boundary;

[0018] Based on ventilation corridor areas, areas with frequent mechanical activity, and dynamically expanding areas, a spatial sampling point layout design is carried out. The spatial sampling point layout design divides the ventilation corridor areas, areas with frequent mechanical activity, and dynamically expanding areas in the remote sensing image into several 5km×5km grid units, and automatically marks at least two sensitive disturbance sampling points that meet preset conditions in each grid unit. The preset conditions include that the vegetation coverage NDVI of the sampling point is less than 0.1, there is a historical construction path nearby, and the terrain is relatively open and free from building and water interference.

[0019] The remote sensing images are unified into a complete remote sensing image sequence based on spatial and temporal labels, consisting of 7 remote sensing images arranged in chronological order. After being relayed and cached, the images are transmitted to the edge computing terminal deployed locally at the ground station via a dedicated link.

[0020] S12. Perform image preprocessing on the remote sensing image sequence in the edge computing terminal to obtain a standardized image set;

[0021] Image preprocessing includes geometric orthorectification, radiometric normalization, cloud masking, and format conversion;

[0022] Geometric orthorectification is performed on all images in a remote sensing image sequence by using the digital elevation model (DEM) and the orbital parameters and attitude control information of the remote sensing platform.

[0023] Radiometric normalization is performed on all images in the geometrically orthorectified remote sensing image sequence. The permanent invariant feature point method (PIF) is used, and a preset reference image is used as the standard. The reflectance of ground objects in images acquired at different times is normalized to unify the spectral response curves of all images in the remote sensing image sequence.

[0024] Cloud masking is performed on all images in the radiometrically normalized remote sensing image sequence. The cloud masking process uses the Fmask algorithm based on pixel-level classification to identify high-reflectivity interference areas, and generates an effective pixel mask after marking cloud and cloud shadow areas.

[0025] The format conversion is achieved by uniformly converting remote sensing image sequences after geometric orthorectification, radiometric normalization, and cloud masking into the TOA (Topographic Aspect) format of land surface reflectance.

[0026] Preferably, S2 includes S21;

[0027] S21. Use OpenCV image processing tools to process all images in the standardized image set into grayscale, obtain grayscale images, sort the grayscale images according to time labels, construct a grayscale image sequence set, and then extract perturbation features from the grayscale image sequence set to obtain an environmental perturbation dataset.

[0028] The environmental disturbance dataset includes the dust space particle expansion diameter Dpe, the dust density heterogeneity factor DnhF, and the environmental wind speed modulation factor Wmf;

[0029] The dust space particle expansion diameter Dpe is obtained by performing gray-level difference processing on two adjacent images within the corresponding area of ​​each sensitive disturbance sampling point to generate a disturbance difference map. In the disturbance difference map, pixels with a disturbance intensity exceeding 10% reflectance difference are screened in the mask area where the NDVI of the base vegetation coverage is less than 0.1, and disturbance patch areas are extracted. After applying closing operation and area filtering to remove small areas for each disturbance patch area, the disturbance boundary contour is extracted using the convex hull algorithm, and the diameter of the smallest circumcircle of the disturbance area is taken as the dust space particle expansion diameter Dpe.

[0030] The dust density heterogeneity factor DnhF is obtained by extracting the set of gray values ​​of all disturbance response pixels in the grid cell of each sensitive disturbance sampling point, calculating the standard deviation of the set of gray values ​​in the sensitive disturbance sampling point region as the texture dispersion within the region, taking the pixels within 5 pixel units outside the outer edge of the disturbance patch as the reference background region, calculating the background gray standard deviation, and then using the standard deviation as the denominator and the background gray standard deviation as the numerator to calculate the ratio.

[0031] The environmental wind speed modulation factor (Wmf) is obtained by launching the API application interface of the meteorological reanalysis data platform to acquire the wind speed time series of several consecutive days before and after the occurrence of the disturbance in the target area. The wind speed time series is then subjected to third-order spline interpolation and moving average fitting to calculate the absolute difference between the wind speed on the day of the disturbance and the average wind speed of the three days before and after that day.

[0032] Preferably, S2 also includes S22;

[0033] S22. Perform data preprocessing on the environmental disturbance dataset. Data preprocessing includes normalization, outlier handling, and sequential indexing.

[0034] Normalization is performed on the environmental disturbance dataset using the Z-score method to normalize the scale and standardize it to a uniform numerical range, thus eliminating the influence of different dimensions between data.

[0035] Outlier handling removes non-physical outliers from environmentally disturbed data by employing a two-sided median filtering algorithm.

[0036] Sequential indexing is achieved by reindexing all environmental disturbance datasets in chronological order and constructing a standardized environmental disturbance dataset.

[0037] Preferably, S3 includes S31;

[0038] S31. Based on the standardized environmental disturbance dataset, the Python environment is used to call NumPy and the image analysis library to extract parameters for each sensitive disturbance sampling point pixel by pixel. The calculation is performed on the edge computing terminal to output the disturbance confidence score index Dti for each sensitive disturbance sampling point, thereby quantifying the degree of retention of disturbance afterimages in remote sensing images after the disturbance behavior occurs.

[0039] The perturbation confidence score index Dti is calculated and output using the following algorithm formula;

[0040]

[0041] In the formula, log e This represents the natural logarithm function.

[0042] Preferably, S3 also includes S32;

[0043] S32. After the calculation is completed, the perturbation confidence score index Dti of each sensitive perturbation sampling point is initially compared and evaluated, and the perturbation confidence level is classified based on the initial comparison and evaluation results. The specific evaluation content is as follows.

[0044] When the perturbation confidence score index Dti < 0.2, it is classified as level 0, indicating a non-perturbation area that does not require attention.

[0045] When 0.2 ≤ Disturbance Confidence Score Index Dti < 0.6, it is classified as Level 1, indicating that the disturbance is questionable. It is automatically marked as a suspicious disturbance area and enters the periodic dynamic re-judgment mechanism. The disturbance confidence score index Dti is continuously tracked for the current sensitive disturbance sampling point in the next 5 remote sensing images. If the disturbance confidence score index Dti is maintained at > 0.5, it will be upgraded to Level 2.

[0046] When the perturbation confidence score index Dti≥0.6, it is divided into 2 levels, indicating a suspected perturbation, and the time spectrum inversion process is triggered.

[0047] Preferably, S4 includes S41;

[0048] S41. When the current sensitive disturbance sampling points are classified into Level 2 in the preliminary comparative evaluation, the time spectrum inversion process is triggered. The time spectrum inversion process extracts 7 remote sensing images in the past 30 days from the suspected sensitive disturbance sampling points and performs image preprocessing to obtain an image time series set.

[0049] For each period of remote sensing images in the image time series set, the perturbation amplitude is extracted using the image differencing method to obtain the perturbation amplitude f of the average pixel in the perturbation region at time phase t. t And the perturbation amplitude f of the average pixel in the perturbation region in the t-th time phase. t Normalization is performed to normalize the disturbance amplitude f to the 0-1 range, eliminating the influence of unit dimensions.

[0050] Preferably, S4 also includes S42;

[0051] S42. The perturbation amplitude f based on the average pixel value of the perturbation region in the t-th time phase. t The calculation is performed, and the perturbation frequency coherence factor AdfC is output to quantify the periodicity and regularity of the perturbation behavior.

[0052] The perturbation frequency coherence factor AdfC is calculated and output using the following algorithm formula;

[0053]

[0054] In the formula, N represents the total number of time phases of the remote sensing image, sin 2 denoted as the sine square function, π represents pi (circular diameter), and is taken to two decimal places. ΔT represents the time interval between remote sensing images.

[0055] Preferably, S5 includes S51;

[0056] S51. Based on the perturbation frequency coherence factor AdfC and the perturbation confidence score index Dti, a joint expression is formed by linear weighted summation to form a spatial and temporal dual-dimensional perturbation feature, and a comprehensive perturbation index Dci is obtained.

[0057] The comprehensive disturbance index Dci is calculated and output using the following algorithm formula;

[0058]

[0059] Preferably, S5 also includes S52;

[0060] S52. The disturbance risk interval threshold includes the first disturbance risk threshold F1 and the second disturbance risk threshold F2. A second comparison evaluation is performed by setting the disturbance risk interval threshold and the comprehensive disturbance index Dci, and the corresponding strategy is executed based on the second comparison evaluation result. The specific evaluation content is as follows.

[0061] When the comprehensive disturbance index Dci is less than the first disturbance risk threshold F1, it is considered as a level 1 disturbance intensity, i.e., a stable background region with insignificant periodic characteristics. It is regarded as a background region and automatically ignored, and does not need to be included in the identification task queue.

[0062] When the first disturbance risk threshold F1 ≤ the comprehensive disturbance index Dci < the second disturbance risk threshold F2, it is represented as a secondary disturbance intensity, i.e. a suspicious disturbance area. Sensitive disturbance sampling points are included in the suspicious layer marking and enter the rolling observation and dynamic update mechanism of the next image.

[0063] When the comprehensive disturbance index Dci is greater than or equal to the second disturbance risk threshold F2, it is considered to be a level 3 disturbance intensity, i.e. a key monitoring area. At this time, it is marked as an abnormal disturbance area, triggering the highest priority monitoring and being included in the priority path of drone patrol routes.

[0064] Among them, the intensity of Level 1 disturbance is less than that of Level 2 disturbance, which is less than that of Level 3 disturbance.

[0065] This invention provides a method for identifying land use disturbed by human activity. It has the following beneficial effects:

[0066] (1) This method constructs a standardized image set by performing unified geometric orthorectification, radiometric normalization, cloud masking and format conversion on remote sensing image sequences in the edge computing terminal. This effectively improves the temporal comparability and spectral consistency between remote sensing images, solves the influencing factors such as illumination differences, geographical offset and cloud interference in multi-temporal images, and provides a high-quality input data foundation for subsequent disturbance feature extraction. At the same time, this invention combines wind vent recognition algorithm, mechanical activity path detection and construction boundary expansion recognition technology to comprehensively extract ventilation corridor area, dynamic expansion area and mechanical disturbance area, and construct a spatial layout scheme of sensitive disturbance sampling points oriented towards disturbance potential. This significantly improves the perception coverage and detection efficiency of key areas, reduces redundant computing resource consumption, and improves the system's real-time processing capability and early recognition accuracy.

[0067] (2) This method introduces three disturbance feature factors: the spatial particle expansion diameter of dust, the dust density heterogeneity factor, DnhF, and the environmental wind speed modulation factor, Wmf. It also innovatively constructs a disturbance credibility scoring index, Dti, to collaboratively model the scale intensity, gray-scale dispersion, and wind speed suppression effect of the disturbance area, effectively distinguishing between real disturbance signals and natural fluctuation background. The disturbance level is classified by the disturbance credibility scoring index Dti, achieving hierarchical classification from level 0 to level 2. Corresponding processing strategies can be formulated according to different levels, such as ignoring, periodic re-judgment, and triggering deep analysis processes, which significantly improves the accuracy, precision, and intelligent response capability of disturbance detection, and solves the technical pain points of "strong signals with low credibility" and "weak signals with high misjudgment" in traditional remote sensing disturbance identification.

[0068] (3) Based on the suspected disturbance area, a time-spectrum inversion process is triggered to construct a time series of disturbance pixel intensity. The disturbance frequency coherence factor AdfC is proposed to measure the regularity and periodicity of disturbance in time series. Furthermore, it is nonlinearly combined with the disturbance credibility score index Dti to propose a comprehensive disturbance index Dci. A spatiotemporal joint disturbance intensity expression system is constructed to achieve comprehensive identification of occasional disturbances, continuous disturbances and periodic disturbances. Based on the comprehensive disturbance index Dci, a disturbance risk interval threshold is set and a secondary comparison evaluation is performed to realize the intelligent hierarchical processing strategy of "eliminating false disturbances, marking suspicious disturbances, and focusing on high-intensity disturbances". This greatly improves the system's robustness in judging complex disturbance behaviors and the efficiency of automatic supervision response, and has significant engineering application value. Attached Figure Description

[0069] Figure 1 This is a schematic diagram illustrating the steps of a method for identifying human-caused land disturbance according to the present invention;

[0070] Figure 2 A schematic diagram of the spatial sampling point layout design;

[0071] Figure 3 This is a diagram showing the overall structure of the remote sensing disturbance identification system. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Example 1

[0074] This invention provides a method for identifying land use disturbed by human activity. Please refer to [link / reference]. Figure 1 , Figure 2 and Figure 3 This includes the following steps:

[0075] S1. Collect remote sensing image sequences of the target area at multiple consecutive time points, and transmit the remote sensing image sequences to the edge computing terminal. Perform image preprocessing on the remote sensing image sequences in the edge computing terminal to obtain a standardized image set.

[0076] S2. Extract perturbation features from each image in the standardized image set to obtain an environmental perturbation dataset, and preprocess the environmental perturbation data to obtain a standardized environmental perturbation dataset.

[0077] S3. Calculate and output the perturbation confidence score index Dti based on the standardized environmental perturbation dataset, and conduct a preliminary comparative evaluation based on the perturbation confidence score index Dt, and classify the perturbation level.

[0078] S4. Based on the preliminary comparative evaluation results, trigger the time spectrum inversion process, extract N periods of remote sensing images of sensitive disturbance sampling points at level 2, obtain the image time series set, perform feature extraction to obtain the disturbance amplitude f, and then calculate and output the disturbance frequency coherence factor AdfC.

[0079] S5. The perturbation frequency coherence factor AdfC and the perturbation confidence score index Dti are calculated together to output the comprehensive perturbation index Dci. The perturbation risk interval threshold is preset and the comprehensive perturbation index Dci is compared and evaluated a second time. The corresponding strategy is executed based on the results of the second comparison and evaluation.

[0080] In this embodiment, the method automatically collects remote sensing images and meteorological data based on the API application interface in step S1, and performs standardized image preprocessing operations in the edge computing terminal to ensure the consistency of remote sensing data in spatial, spectral, and temporal dimensions, thus establishing a unified high-quality input for subsequent processing. Secondly, in step S2, image grayscale processing and differential algorithms are introduced to construct non-traditional disturbance features such as disturbance spatial particle size factor, grayscale discrete factor, and wind speed modulation factor, forming an environmental disturbance dataset. A standardized disturbance feature matrix is ​​then constructed through normalization and anomaly removal, significantly improving the physical interpretability and expressive accuracy of disturbance modeling. In step S3, an innovative disturbance credibility score index (Dti) is constructed, integrating disturbance intensity, heterogeneity, and wind field disturbance effects. This achieves, for the first time, a quantitative assessment of the "credibility" of disturbance behavior. Furthermore, combined with the Dti classification rules, a dynamic discrimination threshold is established to automatically trigger the spectral inversion process for the secondary disturbance region. In step S4, a disturbance time series is constructed and the disturbance frequency coherence factor AdfC is output to identify periodic disturbance patterns, such as construction round trips and transportation repetitive disturbances, effectively distinguishing random noise from substantial disturbance signals. Finally, in step S5, a comprehensive disturbance index Dci is constructed to achieve a two-dimensional fusion expression of the spatial amplitude characteristics and temporal regularity of the disturbance, and a multi-level disturbance risk interval threshold is introduced to achieve secondary discrimination of disturbance intensity and strategy execution, automatically distinguishing between "background stable zone", "suspicious disturbance zone" and "key monitoring zone", realizing a differentiated response mechanism for practical applications.

[0081] Example 2

[0082] Please see Figure 1 , Figure 2 and Figure 3 Specifically: S1 includes S11 and S12;

[0083] S11. By setting up API application interfaces, connect to the remote sensing satellite data service interface and the ERA5 meteorological analysis data platform respectively, and set the acquisition conditions for the API application interfaces to acquire remote sensing images of the target area in real time.

[0084] The acquisition conditions include acquiring multi-temporal image sequences over the past 60 days, with the images being captured at 12:00 ± 1 hour each day and cloud cover less than 10%, and the image bands covering the visible and near-infrared bands.

[0085] The wind corridor effect is analyzed in remote sensing images using a wind vent recognition algorithm to extract ventilation corridor areas;

[0086] A remote sensing image automatic rut identification algorithm is used to locate areas with frequent mechanical activity.

[0087] Edge change detection is performed by comparing remote sensing images from each period to identify the dynamic expansion area of ​​the construction boundary;

[0088] Based on ventilation corridor areas, areas with frequent mechanical activity, and dynamically expanding areas, a spatial sampling point layout design is carried out. The spatial sampling point layout design divides the ventilation corridor areas, areas with frequent mechanical activity, and dynamically expanding areas in the remote sensing image into several 5km×5km grid units, and automatically marks at least two sensitive disturbance sampling points that meet preset conditions in each grid unit. The preset conditions include that the vegetation coverage NDVI of the sampling point is less than 0.1, there is a historical construction path nearby, and the terrain is relatively open and free from building and water interference.

[0089] The remote sensing images are unified into a complete remote sensing image sequence based on spatial and temporal labels, consisting of 7 remote sensing images arranged in chronological order. After being relayed and cached, the images are transmitted to the edge computing terminal deployed locally at the ground station via a dedicated link.

[0090] S12. Perform image preprocessing on the remote sensing image sequence in the edge computing terminal to obtain a standardized image set;

[0091] Image preprocessing includes geometric orthorectification, radiometric normalization, cloud masking, and format conversion;

[0092] Geometric orthorectification uses the digital elevation model (DEM) and the orbital parameters and attitude control information of the remote sensing platform to perform orthorectification on all images in the remote sensing image sequence, so that the images have spatial consistency and eliminate the geographical offset error between images acquired at different times.

[0093] Radiometric normalization is performed on all images in the geometrically orthorectified remote sensing image sequence. The permanent invariant feature point method (PIF) is used, and a preset reference image is used as the standard. The reflectance of ground objects in images acquired at different times is normalized to unify the spectral response curves of all images in the remote sensing image sequence.

[0094] Cloud masking is performed on all images in the radiometrically normalized remote sensing image sequence. The cloud masking process uses the Fmask algorithm based on pixel-level classification to identify high-reflectivity interference areas, and generates an effective pixel mask after marking cloud and cloud shadow areas, which is used for subsequent disturbance detection area screening.

[0095] The format conversion transforms remote sensing image sequences after geometric orthorectification, radiometric normalization, and cloud masking into a unified TOA (Topographic Array) format for easy texture and spectral analysis.

[0096] In this embodiment, the method connects remote sensing satellite data services and the ERA5 meteorological reanalysis platform via an API application programming interface to automatically acquire multi-temporal remote sensing images and meteorological data of the target area over the past 60 days, ensuring data continuity and spatiotemporal coordination. By setting acquisition conditions such as shooting time, cloud cover limit, and band coverage, the impact of low-quality images on disturbance identification accuracy is effectively avoided. Simultaneously, pre-screening algorithms such as wind corridor effect analysis, rut identification, and dynamic monitoring of construction boundaries are introduced to accurately extract ventilation corridor areas, areas with frequent mechanical activity, and dynamically expanding areas. Based on the aforementioned high-risk disturbance areas, a 5km×5km grid is divided and sensitive disturbance sampling points are automatically laid out, ensuring the representativeness, typicality, and monitoring value of the spatial samples, providing high-quality initial samples for spatial modeling of disturbance behavior. For common error factors in remote sensing images, such as geometric, radiometric, and cloud interference, geometric orthorectification, radiometric normalization, cloud masking, and format conversion operations are performed sequentially. Geometric orthorectification calibrates the spatial consistency of images based on the DEM and the attitude parameters of the remote sensing platform; radiometric normalization uses the Permanent Invariant Feature Point (PIF) method to standardize the spectral reflectance of multi-temporal images; cloud masking uses the Fmask algorithm to automatically identify and mask cloud layers and cloud shadow areas, improving effective pixel coverage; finally, the processed images are uniformly converted to the surface reflectance TOA format to ensure the spectral consistency and comparability of remote sensing images in subsequent disturbance intensity extraction and texture feature analysis. In summary, the data acquisition and preprocessing technology system constructed in embodiments S11 and S12 of this invention has three core advantages: "high-precision control at the source, automated process execution, and standardized image output." It solves the problems of scattered remote sensing data sources, unstable quality, and coarse sample layout in traditional methods, significantly improving the timeliness, stability, and interference suppression capability of the entire disturbance identification system, laying the foundation for high-quality, high-confidence disturbance identification.

[0097] Example 3

[0098] Please see Figure 1 Specifically: S2 includes S21;

[0099] S21. Use OpenCV image processing tools to process all images in the standardized image set into grayscale, obtain grayscale images, sort the grayscale images according to time labels, construct a grayscale image sequence set, and then extract perturbation features from the grayscale image sequence set to obtain an environmental perturbation dataset.

[0100] The environmental disturbance dataset includes the dust space particle expansion diameter Dpe, the dust density heterogeneity factor DnhF, and the environmental wind speed modulation factor Wmf;

[0101] The dust space particle expansion diameter Dpe is obtained by performing gray-level difference processing on two adjacent images within the corresponding area of ​​each sensitive disturbance sampling point to generate a disturbance difference map. In the disturbance difference map, pixels with a disturbance intensity exceeding 10% reflectance difference are screened in the mask area where the NDVI of the base vegetation coverage is less than 0.1, and disturbance patch areas are extracted. After applying closing operation and area filtering to remove small areas for each disturbance patch area, the disturbance boundary contour is extracted using the convex hull algorithm, and the diameter of the smallest circumcircle of the disturbance area is taken as the dust space particle expansion diameter Dpe.

[0102] The dust density heterogeneity factor DnhF is obtained by extracting the set of gray values ​​of all disturbance response pixels in the grid cell of each sensitive disturbance sampling point, calculating the standard deviation of the set of gray values ​​in the sensitive disturbance sampling point region as the texture dispersion within the region, taking the pixels within 5 pixel units outside the outer edge of the disturbance patch as the reference background region, calculating the background gray standard deviation, and then using the standard deviation as the denominator and the background gray standard deviation as the numerator to calculate the ratio.

[0103] The environmental wind speed modulation factor (Wmf) is obtained by launching the API application interface of the meteorological reanalysis data platform to acquire the wind speed time series of several consecutive days before and after the occurrence of the disturbance in the target area. The wind speed time series is then subjected to third-order spline interpolation and moving average fitting to calculate the absolute difference between the wind speed on the day of the disturbance and the average wind speed of the three days before and after that day.

[0104] S2 also includes S22;

[0105] S22. Perform data preprocessing on the environmental disturbance dataset. Data preprocessing includes normalization, outlier handling, and sequential indexing.

[0106] Normalization is performed on the environmental disturbance dataset using the Z-score method to normalize the scale and standardize it to a uniform numerical range, thus eliminating the influence of different dimensions between data.

[0107] Outlier handling removes non-physical outliers from environmentally disturbed data by employing a two-sided median filtering algorithm.

[0108] Sequential indexing is achieved by reindexing all environmental disturbance datasets in chronological order and constructing a standardized environmental disturbance dataset.

[0109] In this embodiment, the method constructs a grayscale image sequence set by using OpenCV image processing tools to perform grayscale conversion and temporal sorting on remote sensing images based on a standardized image set, thereby providing a unified processing object for multi-temporal disturbance feature extraction. Through grayscale difference analysis and mask screening operations, pixel patches with strong disturbance responses in sparse vegetation areas are extracted. Combined with closing operations, area filtering, and convex hull contour extraction techniques, the disturbance expansion region is accurately identified and its minimum circumcircle diameter is calculated to obtain the particle expansion diameter Dpe in the dust space, effectively characterizing the spatial scale and outward expansion trend of the disturbance. Simultaneously, by statistically analyzing the grayscale dispersion of disturbance response pixels and comparing it with the surrounding reference area to calculate the ratio, the dust density heterogeneity factor DnhF is obtained, reflecting the internal structural complexity of the disturbance from a texture perspective. Furthermore, the environmental wind speed modulation factor Wmf, based on the ERA5 meteorological platform API, combines third-order spline interpolation and moving average fitting of wind speed time series to extract the difference between the wind speed on the day of the disturbance and the average wind speed of surrounding dates, reflecting the magnitude of the impact of wind field disturbance on image feature changes, thus introducing a disturbance compensation factor from a meteorological perspective. Further, the above three types of disturbance index datasets undergo data standardization and optimization. The Z-score normalization method unifies the numerical scale of each disturbance factor, eliminating unit and distribution differences; two-sided median filtering is used to remove non-physical outliers introduced by remote sensing noise or meteorological anomalies, improving the stability and accuracy of the indicators; finally, sequential indexing is used to reconstruct each indicator according to temporal logic, constructing a standardized environmental disturbance dataset, providing a highly consistent and reliable basic data source for the calculation and classification of the disturbance credibility score index Dti. This invention, through the disturbance feature construction and standardized data processing technology described in step S2, not only achieves a systematic characterization of the spatial structure, texture complexity, and meteorological modulation effect of disturbances, but also significantly improves the robustness and comparability of indicators through the data preprocessing process, enhances the accuracy and anti-interference ability of subsequent disturbance behavior analysis, and effectively solves the technical bottleneck problems such as "unstable image features", "unmodeled wind field influence" and "blurred disturbance boundaries" in regional remote sensing disturbance identification.

[0110] Example 4

[0111] Please see Figure 1Specifically: S3 includes S31;

[0112] S31. Based on the standardized environmental disturbance dataset, the Python environment is used to call NumPy and the image analysis library to extract parameters for each sensitive disturbance sampling point pixel by pixel. The calculation is performed on the edge computing terminal to output the disturbance confidence score index Dti for each sensitive disturbance sampling point, thereby quantifying the degree of retention of disturbance afterimages in remote sensing images after the disturbance behavior occurs.

[0113] The perturbation confidence score index Dti is calculated and output using the following algorithm formula;

[0114]

[0115] In the formula, log e Represent the natural logarithm function;

[0116] The calculation logic and derivation process of the formula: The numerator is used to measure the spatial scale and internal perturbation intensity of the perturbation region. The two together improve the credibility score and simulate the positive contribution of perturbation significance and perturbation complexity to the score.

[0117] The denominator represents the impact of wind speed. A suppression term is introduced, using a logarithmic function to compress the noise effect of high wind speeds and prevent excessive reduction. Adding 1 avoids a denominator of 0. Wmf 2 Used to enhance sensitivity to the effects of high wind speeds;

[0118] Modeling principles for the challenge of perturbation afterimage recognition: Wind field changes are the main interference factors in perturbation information recognition, namely blurred boundaries, displacement, and masking; Image difference can reveal spatially abrupt change regions, but high-frequency texture and gray-level inhomogeneity are the real perturbation signals;

[0119] The construction principle of the multi-factor scoring system is as follows: the particle expansion diameter Dpe and the dust density heterogeneity factor DnhF are perturbation enhancement terms and need to be positively coupled; the environmental wind speed modulation factor Wmf is a negative factor and must decay nonlinearly to prevent high wind speed areas from being excluded.

[0120] The approach to dimensionless parameter normalization is as follows: all parameters involved in the calculation are normalized environmental perturbation datasets that have eliminated the influence of dimensions between data. The calculation result of the perturbation confidence score index Dti is also a dimensionless parameter.

[0121] S3 also includes S32;

[0122] S32. After the calculation is completed, the perturbation confidence score index Dti of each sensitive perturbation sampling point is initially compared and evaluated, and the perturbation confidence level is classified based on the initial comparison and evaluation results. The specific evaluation content is as follows.

[0123] When the perturbation confidence score index Dti < 0.2, it is classified as level 0, indicating a non-perturbation area that does not require attention.

[0124] When 0.2 ≤ Disturbance Confidence Score Index Dti < 0.6, it is classified as Level 1, indicating that the disturbance is questionable. It is automatically marked as a suspicious disturbance area and enters the periodic dynamic re-judgment mechanism. The disturbance confidence score index Dti is continuously tracked for the current sensitive disturbance sampling point in the next 5 remote sensing images. If the disturbance confidence score index Dti is maintained at > 0.5, it will be upgraded to Level 2.

[0125] When the perturbation confidence score index Dti≥0.6, it is divided into 2 levels, indicating a suspected perturbation, and the time spectrum inversion process is triggered.

[0126] In this embodiment, the method uses a standardized environmental disturbance dataset obtained in the preceding steps. A Python programming environment combined with NumPy and an image analysis library is used on an edge computing terminal to extract parameters pixel-by-pixel from each sensitive disturbance sampling point. The disturbance credibility scoring index (Dti) is then calculated in real-time using the Dti algorithm model to form a high-resolution disturbance credibility scoring layer. The Dti comprehensively considers the positive enhancement effects of the dust particle expansion diameter (Dpe) and the dust density heterogeneity factor (DnhF), while simultaneously introducing a negative suppression mechanism using the environmental wind speed modulation factor (Wmf). A nonlinear logarithmic function model is used to suppress and model wind field disturbances, preventing image artifacts caused by high wind speeds from affecting the accuracy of the disturbance score, thus achieving a multi-dimensional evaluation of the disturbance retention degree. Further, the Dti is used for preliminary comparative evaluation and disturbance level classification, setting a grading mechanism from level 0 to level 2 to achieve step-by-step identification and response of disturbance information. This triggers the subsequent time-spectrum inversion process, initiating a higher-level disturbance identification strategy. In summary, this step not only constructs a disturbance scoring index system based on multi-factor fusion and applies it to the multi-level disturbance credibility identification process, but also realizes the quantitative judgment and intelligent hierarchical screening of the credibility of the disturbance area, effectively improving the identification sensitivity, dynamic response capability and spatial positioning accuracy under complex disturbance backgrounds, and significantly enhancing the intelligence level and precise handling capability of the entire disturbance monitoring system.

[0127] Example 5

[0128] Please see Figure 1 Specifically: S4 includes S41;

[0129] S41. When the current sensitive disturbance sampling points are classified into Level 2 in the preliminary comparative evaluation, the time spectrum inversion process is triggered. The time spectrum inversion process extracts 7 remote sensing images in the past 30 days from the suspected sensitive disturbance sampling points and performs image preprocessing to obtain an image time series set.

[0130] For each period of remote sensing images in the image time series set, the perturbation amplitude is extracted using the image differencing method to obtain the perturbation amplitude f of the average pixel in the perturbation region at time phase t. t And the perturbation amplitude f of the average pixel in the perturbation region in the t-th time phase. t Normalization is performed to normalize the disturbance amplitude f to the 0-1 range, eliminating the influence of unit dimensions.

[0131] S4 also includes S42;

[0132] S42. The perturbation amplitude f based on the average pixel value of the perturbation region in the t-th time phase. t The calculation is performed, and the perturbation frequency coherence factor AdfC is output to quantify the periodicity and regularity of the perturbation behavior.

[0133] The perturbation frequency coherence factor AdfC is calculated and output using the following algorithm formula;

[0134]

[0135] In the formula, N represents the total number of time phases of the remote sensing image, sin 2 denoted as the sine square function, π represents the value of pi, taken to two decimal places, and ΔT represents the time interval of the remote sensing image.

[0136] The background logic of the formula: Under the influence of periodic human activities such as construction and transportation or natural disturbances such as wind erosion, the disturbances in the target area may recur in an intermittent or regular manner. If the disturbance shows a "rhythmic" fluctuation in the remote sensing image sequence, it can be regarded as a non-random event and needs to be included in the key identification strategy.

[0137] The formula is derived based on the following: Fourier analysis is used to simulate the periodicity of the disturbance with a simple sine function; the amplitude value f of each time-phase disturbance is treated as a frequency amplitude term and mapped to sin... 2 (2π·f t • △T), forming a periodic coupled response; finally, mean aggregation is used to calculate the degree of concentration of the periodic response to the disturbance;

[0138] sine square function sin 2 The function of the sine square function: If the disturbance is a high-frequency fluctuation, then the sine square function sin 2 The response fluctuates significantly; if the disturbance is a regular periodic change, then the sine square function sin 2 The response clusters close to 1; if the perturbation is random or noise interference, the response value does not show obvious clustering between [0,1].

[0139] In this embodiment, the method extracts seven remote sensing images of the corresponding sensitive disturbance sampling points within the past 30 days, performs image preprocessing, and constructs an image time series set, thus achieving continuous observation of the disturbance area in the time dimension. Subsequently, the image difference method is used to extract the disturbance amplitude information from each period of the image, and the disturbance amplitude f of the average pixel in the disturbance area in the t-th time phase is calculated. t Furthermore, by normalizing the scale, the comparability of perturbation amplitudes f at different time points is ensured in subsequent analyses, eliminating interference from different unit dimensions or changes in image brightness. Based on the above perturbation amplitude sequence, a perturbation frequency coherence factor AdfC is constructed. A periodic response model with a sinusoidal square function as its core is used, combined with Fourier analysis, to simulate the temporal fluctuations of perturbation behavior. This perturbation frequency coherence factor AdfC achieves a quantitative assessment of the degree of perturbation periodicity by mapping the percussion amplitude value f of each period to a sinusoidal periodicity: when the perturbation exhibits periodic fluctuations, the sinusoidal square function sin... 2 The response tends to be concentrated, and the perturbation frequency coherence factor AdfC value increases significantly; when the perturbation is an occasional or random perturbation, sin 2 The response volatility increases, while the perturbation frequency coherence factor AdfC value tends to disperse. This mechanism can effectively capture the "rhythmic" perturbation characteristics manifested by intermittent human activities or natural disturbances such as construction and transportation, thereby achieving an upgraded identification of key perturbation areas. In summary, through this implementation method, the present invention not only realizes three-dimensional modeling from spatial perturbation to temporal perturbation and constructs a linkage mechanism between perturbation credibility scoring and perturbation frequency response, but also significantly improves the sensitivity and stability of identifying periodic perturbation behavior by introducing the perturbation frequency coherence factor AdfC. This provides precise support for subsequent perturbation level reassessment and disposal strategy optimization, achieving the goal of leaping from "static visibility" to "dynamic judgment" in perturbation perception.

[0140] Example 6

[0141] Please see Figure 1 Specifically: S5 includes S51;

[0142] S51. Based on the perturbation frequency coherence factor AdfC and the perturbation confidence score index Dti, a joint expression is formed by linear weighted summation to form a spatial and temporal dual-dimensional perturbation feature, and a comprehensive perturbation index Dci is obtained.

[0143] The comprehensive disturbance index Dci is calculated and output using the following algorithm formula;

[0144]

[0145] The derivation logic of the formula: The product structure ensures that the perturbation confidence score index Dti is the dominant variable, and the perturbation frequency coherence factor AdfC plays a modulation role. Only when both are high at the same time will the perturbation confidence score index Dti increase significantly, which is more in line with the actual remote sensing perturbation scenario.

[0146] Linear weighted summation can lead to the periodic amplification of weak disturbances. That is, when the disturbance confidence score index Dti is low but the disturbance frequency coherence factor AdfC is high, there is still a high comprehensive disturbance index Dci. The square root is used to make the disturbance frequency coherence factor AdfC too sensitive to changes, and loses the dominance of disturbance confidence.

[0147] S5 also includes S52;

[0148] S52. The disturbance risk interval threshold includes a first disturbance risk threshold F1 and a second disturbance risk threshold F2; wherein the first disturbance risk threshold F1 serves as the effective lower limit threshold for eliminating false disturbances, and the second disturbance risk threshold F2 serves as the activation threshold for marking high-intensity disturbances; a second comparison evaluation is performed by using the preset disturbance risk interval threshold and the comprehensive disturbance index Dci, and the corresponding strategy is executed based on the results of the second comparison evaluation. The specific evaluation content is as follows.

[0149] When the comprehensive disturbance index Dci is less than the first disturbance risk threshold F1, it is considered as a level 1 disturbance intensity, i.e., a stable background region with insignificant periodic characteristics. It is regarded as a background region and automatically ignored, and does not need to be included in the identification task queue.

[0150] When the first disturbance risk threshold F1 ≤ the comprehensive disturbance index Dci < the second disturbance risk threshold F2, it is represented as a secondary disturbance intensity, i.e. a suspicious disturbance area. Sensitive disturbance sampling points are included in the suspicious layer marking and enter the rolling observation and dynamic update mechanism of the next image.

[0151] When the comprehensive disturbance index Dci is greater than or equal to the second disturbance risk threshold F2, it is considered to be a level 3 disturbance intensity, i.e. a key monitoring area. At this time, it is marked as an abnormal disturbance area, triggering the highest priority monitoring and being included in the priority path of drone patrol routes.

[0152] Among them, the intensity of Level 1 disturbance is less than that of Level 2 disturbance, which is less than that of Level 3 disturbance.

[0153] In this embodiment, the method constructs a product-type expression structure with the perturbation confidence score index Dti as the dominant factor and the perturbation frequency coherence factor AdfC as the adjustment term. This effectively avoids the false alarm problem caused by the amplification of weak perturbation risk due to periodic features in the linear weighted model. The comprehensive perturbation index Dci reflects the coupling characteristics of perturbation intensity and perturbation periodicity in the form of a structural function. It only generates a high-intensity perturbation score when both increase simultaneously, which is more in line with the real physical scenario of "perturbation afterimage and perturbation pattern" in remote sensing perturbation identification, thereby improving the accuracy and robustness of abnormal perturbation identification. By setting perturbation risk interval thresholds, a first perturbation risk threshold F1 and a second perturbation risk threshold F2 are defined respectively, and the comprehensive perturbation index Dci is evaluated twice based on these thresholds to achieve further refinement and dynamic classification of perturbation levels. By setting F1 as the false disturbance rejection threshold, low-reliability responses caused by instantaneous wind disturbances, sensor errors, or low-frequency construction activities can be effectively filtered out. By setting F2 as the high disturbance activation threshold, areas of sudden large-scale construction or long-term abnormal disturbances can be accurately captured. In summary, through this implementation method, the present invention achieves full-process linkage modeling from disturbance intensity assessment to disturbance periodicity identification, and introduces an adjustable disturbance risk classification strategy. This not only improves the scientific nature and accuracy of disturbance identification, but also provides a highly reliable decision-making basis for the intelligent and differentiated response of regulatory scheduling strategies, ultimately achieving the remote sensing disturbance identification goal of "accurate identification, real-time response, and intelligent linkage".

[0154] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying land use disturbed by human activity, characterized in that: Includes the following steps: S1. Collect remote sensing image sequences of the target area at multiple consecutive time points, and transmit the remote sensing image sequences to the edge computing terminal. Perform image preprocessing on the remote sensing image sequences in the edge computing terminal to obtain a standardized image set. S2. Extract perturbation features from each image in the standardized image set to obtain an environmental perturbation dataset, and preprocess the environmental perturbation data to obtain a standardized environmental perturbation dataset. S3. Calculate and output the perturbation confidence score index Dti based on the standardized environmental perturbation dataset, and conduct a preliminary comparative evaluation based on the perturbation confidence score index Dt, and classify the perturbation level. S4. Based on the preliminary comparative evaluation results, trigger the time spectrum inversion process, extract N periods of remote sensing images of sensitive disturbance sampling points at level 2, obtain the image time series set, perform feature extraction to obtain the disturbance amplitude f, and then calculate and output the disturbance frequency coherence factor AdfC. S5. The perturbation frequency coherence factor AdfC and the perturbation confidence score index Dti are calculated together to output the comprehensive perturbation index Dci. The perturbation risk interval threshold is preset and the comprehensive perturbation index Dci is compared and evaluated a second time. The corresponding strategy is executed based on the results of the second comparison and evaluation.

2. The method for identifying land use disturbed by human activity according to claim 1, characterized in that: S1 includes S11 and S12; S11. By setting up API application interfaces, connect to the remote sensing satellite data service interface and the ERA5 meteorological analysis data platform respectively, and set the acquisition conditions for the API application interfaces to acquire remote sensing images of the target area in real time. The acquisition conditions include acquiring multi-temporal image sequences over the past 60 days, with the image capture time being within 12:00 ± 1 hour of noon each day and the cloud cover being less than 10%, and the image wavelength range covering the visible and near-infrared bands. The wind corridor effect is analyzed in remote sensing images using a wind vent recognition algorithm to extract ventilation corridor areas; A remote sensing image automatic rut identification algorithm is used to locate areas with frequent mechanical activity. Edge change detection is performed by comparing remote sensing images from each period to identify the dynamic expansion area of ​​the construction boundary; Based on ventilation corridor areas, areas with frequent mechanical activity, and dynamically expanding areas, a spatial sampling point layout design is carried out. The spatial sampling point layout design divides the ventilation corridor areas, areas with frequent mechanical activity, and dynamically expanding areas in the remote sensing image into several 5km×5km grid units, and automatically marks at least two sensitive disturbance sampling points that meet preset conditions in each grid unit. The preset conditions include that the vegetation coverage NDVI of the sampling point is less than 0.1 and that there is a historical construction path nearby. The remote sensing images are unified into a complete remote sensing image sequence based on spatial and temporal labels, consisting of 7 remote sensing images arranged in chronological order. After being relayed and cached, the images are transmitted to the edge computing terminal deployed locally at the ground station via a dedicated link. S12. Perform image preprocessing on the remote sensing image sequence in the edge computing terminal to obtain a standardized image set; The image preprocessing includes geometric orthorectification, radiometric normalization, cloud masking, and format conversion. The geometric orthorectification is performed on all images in the remote sensing image sequence by using the digital elevation model (DEM) and the orbital parameters and attitude control information of the remote sensing platform. The radiometric normalization is performed on all images in the geometrically orthorectified remote sensing image sequence. The permanent invariant feature point (PIF) method is used, and a preset reference image is used as the standard. The reflectance of ground objects in images acquired at different times is normalized to unify the spectral response curves of all images in the remote sensing image sequence. The cloud masking process is performed on all images in the radiometrically normalized remote sensing image sequence. The cloud masking process uses the Fmask algorithm based on pixel-level classification to identify high-reflectivity interference areas, and generates an effective pixel mask after marking the cloud layer and cloud shadow areas. The format conversion involves uniformly converting remote sensing image sequences after geometric orthorectification, radiometric normalization, and cloud masking into the TOA (Total Earth Reflectance) format.

3. The method for identifying land use disturbed by human activity according to claim 1, characterized in that: S2 includes S21; S21. Use OpenCV image processing tools to process all images in the standardized image set into grayscale, obtain grayscale images, sort the grayscale images according to time labels, construct a grayscale image sequence set, and then extract perturbation features from the grayscale image sequence set to obtain an environmental perturbation dataset. The environmental disturbance dataset includes the dust space particle expansion diameter Dpe, the dust density heterogeneity factor DnhF, and the environmental wind speed modulation factor Wmf. The dust space particle expansion diameter Dpe is obtained by performing gray-level difference processing on two adjacent images within the corresponding area of ​​each sensitive disturbance sampling point to generate a disturbance difference map. In the disturbance difference map, pixels with a disturbance intensity exceeding 10% reflectance difference are screened in the mask area where the NDVI of the base vegetation coverage is less than 0.1, and disturbance patch areas are extracted. After applying closing operation and area filtering to remove small areas for each disturbance patch area, the disturbance boundary contour is extracted using the convex hull algorithm, and the diameter of the smallest circumcircle of the disturbance area is taken as the dust space particle expansion diameter Dpe. The dust density heterogeneity factor DnhF is obtained by extracting the set of gray values ​​of all disturbance response pixels in the grid cell of each sensitive disturbance sampling point, calculating the standard deviation of the set of gray values ​​in the sensitive disturbance sampling point region as the texture dispersion within the region, taking the pixels within 5 pixel units of the outer edge of the disturbance patch as the reference background region, calculating the background gray standard deviation, and then using the standard deviation as the denominator and the background gray standard deviation as the numerator to perform ratio calculation. The environmental wind speed modulation factor Wmf is obtained by activating the API application interface of the meteorological reanalysis data platform to acquire the wind speed time series of several consecutive days before and after the disturbance in the target area. The wind speed time series is then subjected to third-order spline interpolation and moving average fitting to calculate the absolute difference between the wind speed on the day of the disturbance and the average wind speed of the three days before and after that day.

4. The method for identifying land use disturbed by human activity according to claim 3, characterized in that: S2 further includes S22; S22. Perform data preprocessing on the environmental disturbance dataset, including normalization, outlier handling, and sequential indexing. The normalization process uses the Z-score method to perform scale normalization on the environmental disturbance dataset, standardizing it to a uniform numerical range and eliminating the influence of different dimensions between data. The outlier handling process employs a two-sided median filtering algorithm to remove non-physical outliers from the environmental disturbance dataset. The sequential indexing process involves reindexing all environmental disturbance datasets in chronological order and constructing a standardized environmental disturbance dataset.

5. The method for identifying land use disturbed by human activity according to claim 4, characterized in that: S3 includes S31; S31. Based on the standardized environmental disturbance dataset, the Python environment is used to call NumPy and image analysis libraries to extract parameters for each sensitive disturbance sampling point pixel by pixel. The calculation is performed on the edge computing terminal to output the disturbance confidence score index Dti for each sensitive disturbance sampling point, thereby quantifying the degree of retention of disturbance afterimages in remote sensing images after the occurrence of disturbance behavior.

6. The method for identifying land use disturbed by human activity according to claim 4, characterized in that: S3 further includes S32; S32. After the calculation is completed, the perturbation confidence score index Dti of each sensitive perturbation sampling point is initially compared and evaluated, and the perturbation confidence level is classified based on the initial comparison and evaluation results. The specific evaluation content is as follows. When the perturbation confidence score index Dti < 0.2, it is classified as level 0, indicating a non-perturbation area that does not require attention. When 0.2 ≤ Disturbance Confidence Score Index Dti < 0.6, it is classified as Level 1, indicating that the disturbance is questionable. It is automatically marked as a suspicious disturbance area and enters the periodic dynamic re-judgment mechanism. The disturbance confidence score index Dti is continuously tracked for the current sensitive disturbance sampling point in the next 5 remote sensing images. If the disturbance confidence score index Dti is maintained at > 0.5, it will be upgraded to Level 2. When the perturbation confidence score index Dti≥0.6, it is divided into 2 levels, indicating a suspected perturbation, and the time spectrum inversion process is triggered.

7. The method for identifying land use disturbed by human activity according to claim 6, characterized in that: S4 includes S41; S41. When the current sensitive disturbance sampling points are classified into Level 2 in the preliminary comparative evaluation, the time spectrum inversion process is triggered. The time spectrum inversion process extracts 7 remote sensing images in the past 30 days from the suspected sensitive disturbance sampling points and performs image preprocessing to obtain an image time series set. For each period of remote sensing images in the image time series set, the perturbation amplitude is extracted using the image differencing method to obtain the perturbation amplitude f of the average pixel in the perturbation region at time phase t. t And the perturbation amplitude f of the average pixel in the perturbation region in the t-th time phase. t Normalization is performed to normalize the disturbance amplitude f to the 0-1 range, eliminating the influence of unit dimensions.

8. The method for identifying land use disturbed by human activity according to claim 7, characterized in that: S4 also includes S42; S42. The perturbation amplitude f based on the average pixel value of the perturbation region in the t-th time phase. t The calculation is performed to output the perturbation frequency coherence factor AdfC, which quantifies the periodicity and regularity of the perturbation behavior.

9. A method for identifying land use disturbed by human activity according to claim 8, characterized in that: S5 includes S51; S51. Based on the perturbation frequency coherence factor AdfC and the perturbation confidence score index Dti, a joint expression is formed by linear weighted summation to form a spatial and temporal dual-dimensional perturbation feature, and a comprehensive perturbation index Dci is obtained.

10. A method for identifying land use disturbed by human activity according to claim 9, characterized in that: S5 also includes S52; S52. The disturbance risk interval threshold includes a first disturbance risk threshold F1 and a second disturbance risk threshold F2. A second comparison evaluation is performed by using the preset disturbance risk interval threshold and the comprehensive disturbance index Dci, and the corresponding strategy is executed based on the second comparison evaluation result. The specific evaluation content is as follows. When the comprehensive disturbance index Dci is less than the first disturbance risk threshold F1, it is considered as a level 1 disturbance intensity, i.e., a stable background region with insignificant periodic characteristics. It is regarded as a background region and automatically ignored, and does not need to be included in the identification task queue. When the first disturbance risk threshold F1 ≤ the comprehensive disturbance index Dci < the second disturbance risk threshold F2, it is represented as a secondary disturbance intensity, i.e. a suspicious disturbance area. Sensitive disturbance sampling points are included in the suspicious layer marking and enter the rolling observation and dynamic update mechanism of the next image. When the comprehensive disturbance index Dci is greater than or equal to the second disturbance risk threshold F2, it is considered to be a level 3 disturbance intensity, i.e. a key monitoring area. At this time, it is marked as an abnormal disturbance area, triggering the highest priority monitoring and being included in the priority path of drone patrol routes. Among them, the intensity of Level 1 disturbance is less than that of Level 2 disturbance, which is less than that of Level 3 disturbance.

Images