Regional change detection method and system based on unmanned aerial vehicle vision

Abstract

The invention discloses a region change detection method and system based on unmanned aerial vehicle vision, and the method comprises the steps: constructing a region change detection model through employing images of the same region in different periods, and enabling the region change detection model to capture a dynamic evolution mode (such as gradual vegetation degradation or sudden building extension) of the surface change; the characteristic limitation of traditional single-time phase or double-time phase comparison is broken through; a double-time-phase image pair is generated through a real-time image based on unmanned aerial vehicle vision, it is ensured that the double-time-phase image pair is analyzed under a unified coordinate system and a spectrum reference, the detection stability is improved, and therefore the anti-jamming capability is improved; through the full-process closed-loop design from data division, model training to real-time detection, in combination with the end-to-end reasoning capability of a deep learning model, the manual intervention demand is reduced, manual threshold setting or feature screening is not needed, change pattern spots are directly output in the test stage, and full automation from data input to result output is achieved.

Description

Technical Field

[0001] This invention relates to the field of natural resource monitoring technology, and in particular to a method and system for detecting regional changes based on UAV vision. Background Technology

[0002] With the rapid development of remote sensing technology, the use of multi-temporal remote sensing images for regional land surface change detection has become a core technical means in fields such as geographic national condition monitoring, urban planning and management, and disaster assessment and early warning. Traditional change detection methods mainly rely on manually designed features (such as spectral indices and texture features) to extract changed areas through techniques such as image difference, ratio, and change vector analysis (CVA). However, such methods have problems such as limited feature expression ability, weak anti-interference ability, and low degree of automation in scenarios with complex land cover types, seasonal change interference, or sensor differences, making it difficult for the change detection accuracy to meet the needs of practical applications. Summary of the Invention

[0003] In view of this, the present invention proposes a region change detection method and system based on UAV vision, which can effectively solve the defects of existing technologies such as limited feature expression ability, weak anti-interference ability and low degree of automation.

[0004] The technical solution of this invention is implemented as follows:

[0005] A method for detecting regional changes based on UAV vision, specifically including:

[0006] Select remote sensing images of the same monitoring area at different time phases, and divide them into training set, validation set and test set based on the remote sensing images at different time phases;

[0007] A regional change detection model is constructed based on the division of the training set, validation set, and test set;

[0008] Acquire real-time images from the drone's vision, and generate dual-temporal image pairs based on the real-time images from the drone's vision;

[0009] The regional change detection model is used to process dual-temporal image pairs to identify areas where changes have occurred on the ground surface in the photographed area, extract change patches, and obtain regional change detection results.

[0010] As a further optional scheme of the UAV vision-based region change detection method, the division of remote sensing images into training, validation, and test sets based on multiple different temporal phases specifically includes:

[0011] The coordinate system and projection of multiple remote sensing images from different time periods are unified to obtain unified remote sensing images.

[0012] The remote sensing images after the unification processing are sliced, and invalid data is removed to form an initial multi-temporal data set;

[0013] The double-temporal data in the initial multi-temporal data set is labeled by using a labeling tool suitable for remote sensing double-temporal change detection, and a labeled data set is obtained.

[0014] According to a preset proportion relationship, the labeled data set is divided into a training set, a validation set and a test set, wherein the training set is used for parameter learning of the model, the validation set is used for model selection and hyperparameter optimization, and the test set is used for evaluating the generalization ability of the model.

[0015] As a further optional solution of the region change detection method based on unmanned aerial vehicle vision, the region change detection model is constructed according to the division of the training set, the validation set and the test set, and specifically includes:

[0016] In the pre-trained Changeformer model, the training set is used for end-to-end model parameter learning, so that the Changeformer model captures the nonlinear mapping relationship between the remote sensing images in the training set and the changes;

[0017] In the model training process, the model selection and hyperparameter optimization are performed based on the validation set, and the hyperparameters include the learning rate, the optimizer and the training round;

[0018] The performance of the trained and optimized model is evaluated using the test set, and the model that passes the performance evaluation is the region change detection model.

[0019] As a further optional solution of the region change detection method based on unmanned aerial vehicle vision, the real-time image of the unmanned aerial vehicle vision is obtained, and a double-temporal image pair is generated based on the real-time image of the unmanned aerial vehicle vision, specifically including:

[0020] The unmanned aerial vehicle aerial photography platform is connected to establish a real-time transmission channel for unmanned aerial vehicle images, and real-time aerial photography images of the target monitoring area returned by the unmanned aerial vehicle are received;

[0021] According to the flight route planning of the unmanned aerial vehicle aerial photography task and the real-time aerial photography image, an orthographic image is generated;

[0022] The historical image of the target monitoring area is obtained;

[0023] According to the orthographic image, the corresponding regional image is cut out from the historical image;

[0024] The cut historical corresponding regional image and the orthographic image are respectively sliced according to a preset training sample picture specification to form a double-temporal image pair.

[0025] As a further optional solution of the unmanned aerial vehicle vision-based regional change detection method, the orthographic image is generated according to the flight route planning of the unmanned aerial vehicle aerial photography task and the real-time aerial photography image, and specifically includes:

[0026] The flight route planning file of the unmanned aerial vehicle aerial photography task is analyzed in real time to obtain the spatial distribution of the flight route, the shooting point position, and the image coverage range information;

[0027] The aerial photography image returned by the unmanned aerial vehicle in real time is subjected to quality inspection to obtain the aerial photography image from which the invalid image is removed;

[0028] According to the shooting point position and the image coverage range in the flight route planning, the aerial photography image from which the invalid image is removed is automatically grouped, and the images covering adjacent regions or having overlapping parts are grouped into one group;

[0029] The grouped aerial photography image is sequentially subjected to orientation, correction, and resampling processing to obtain the oriented, corrected, and resampled image;

[0030] The oriented, corrected, and resampled image is spliced according to the grouping to generate the orthographic image covering the target region.

[0031] As a further optional solution of the unmanned aerial vehicle vision-based regional change detection method, the historical image of the target monitoring region is obtained, and specifically includes:

[0032] According to the geographical features of the target monitoring region and the time span requirement of the historical image, the historical image source meeting the condition is screened from a remote sensing image database, a geographic information data sharing platform, or a related historical image archiving agency;

[0033] The projection system and the coordinate system of the historical image are checked, and if they are inconsistent with the preset or currently used projection and coordinate system, projection conversion and coordinate registration processing are performed;

[0034] The checked and processed historical image is stored in a local or designated server.

[0035] As a further optional solution of the unmanned aerial vehicle vision-based regional change detection method, the dual-time-phase image pair is processed based on the regional change detection model to identify the region where the ground surface has changed, extract the change graph, and obtain the regional change detection result, and specifically includes:

[0036] The dual-time-phase image pair is input into the regional change detection model, and the regional change detection model judges whether the ground surface has changed based on the dual-time-phase image pair;

[0037] The region determined to have changed is outlined, and the change graph is extracted;

[0038] According to the image features and the preset change type classification standard, the change type to which each change patch belongs is determined, and the area of each change patch is calculated.

[0039] The change patch and the related attribute information are output in the form of a vector layer.

[0040] A regional change detection system based on unmanned aerial vehicle vision, comprising:

[0041] A data acquisition module is configured to select multiple remote sensing images of different time phases of the same monitoring region.

[0042] A data division module is configured to divide a training set, a verification set and a test set based on the multiple remote sensing images of different time phases.

[0043] A model construction module is configured to construct a regional change detection model according to the division of the training set, the verification set and the test set.

[0044] A real-time image processing module is configured to acquire real-time images of unmanned aerial vehicle vision, and generate a pair of double-time-phase image pictures based on the real-time images of unmanned aerial vehicle vision.

[0045] A change detection module is configured to process the pair of double-time-phase image pictures based on the regional change detection model, identify the region of the ground surface in the camera area where changes have occurred, extract change patches, and obtain a regional change detection result.

[0046] A computing device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of any one of the above-mentioned regional change detection methods based on unmanned aerial vehicle vision.

[0047] A computer-readable storage medium, the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of any one of the above-mentioned regional change detection methods based on unmanned aerial vehicle vision.

[0048] The beneficial effects of the present application are: by using images of different periods of the same region to construct a regional change detection model, the regional change detection model can capture the dynamic evolution pattern of the ground surface changes (such as gradual vegetation degradation or sudden building expansion), breaking through the feature limitations of traditional single-time-phase or double-time-phase comparison; by generating a pair of double-time-phase image pictures based on real-time images of unmanned aerial vehicle vision, ensuring that the pair of double-time-phase images are analyzed under a unified coordinate system and spectral reference, improving detection stability, and thus improving anti-interference capability; through the whole-process closed-loop design from data division, model training to real-time detection, combined with the end-to-end inference capability of the deep learning model, the need for manual intervention is reduced, without the need for manual threshold setting or feature screening, directly outputting change patches in the test stage, realizing full automation from data input to result output. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0050] Figure 1 A flowchart of a regional change detection method based on unmanned aerial vehicle vision according to the present application;

[0051] Figure 2 A composition diagram of a regional change detection system based on unmanned aerial vehicle vision according to the present application;

[0052] Figure 3 A composition diagram of a computing device according to the present application;

[0053] Figure 4 A flowchart of constructing a regional change detection model according to the present application;

[0054] Figure 5 A flowchart of real-time regional change detection according to the present application. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0056] Reference Figures 1 to 5 A regional change detection method based on unmanned aerial vehicle vision, specifically comprising:

[0057] Selecting multiple remote sensing images of different time phases of the same monitoring area, dividing a training set, a verification set and a test set based on the multiple remote sensing images of different time phases; in some embodiments, the dividing of the training set, the verification set and the test set based on the multiple remote sensing images of different time phases specifically comprises:

[0058] Performing a unified processing of coordinate system and projection on the multiple remote sensing images of different time phases to obtain the remote sensing images after the unified processing;

[0059] Performing a slicing operation on the remote sensing images after the unified processing and removing invalid data to form an initial multi-time phase data set;

[0060] The double-phase data is labeled using a labeling tool LabCD, and the data that does not meet the requirements, such as image black edges, poor quality, blurring, obvious shadows, urban green belts and small green areas, is removed. The labeling results are divided into a training set, a validation set and a test set according to a ratio of 7:3:1, wherein the training set is used for parameter learning of the model, the validation set is used for model selection and hyperparameter optimization, and the test set is used for evaluating the generalization ability of the model.

[0061] Specifically, through coordinate system and projection unification processing, geometric distortion caused by coordinate system or projection difference of multi-source remote sensing images is eliminated, strict alignment of different time images on the spatial reference is ensured, false change mis-detection caused by geometric misalignment is avoided, and a foundation is laid for accurate change analysis; large-scale images are split into local calculation units through slicing, the calculation complexity is reduced and the processing efficiency is improved; invalid data such as black edges, poor quality and blurring are removed, noise interference is reduced, and the data purity of subsequent labeling and model training is improved.

[0062] The double-phase images are labeled using a standardization tool to ensure accurate and uniform labeling boundaries. By removing image black edges, poor quality areas and pseudo changes (such as green belt increase and decrease), noise during labeling is reduced, and the signal-to-noise ratio of training data is improved. The training set, validation set and test set are divided according to a ratio of 7:3:1, taking into account the needs of model learning, hyperparameter optimization and performance evaluation. The training set covers sufficient samples to learn change patterns, the validation set optimizes the model structure to prevent overfitting, and the test set independently evaluates the generalization ability to ensure the stability of the model in real scenarios.

[0063] Through multi-period image comparison, the model can capture the dynamic process of surface changes (such as gradual land degradation or phased urban construction), extract time-related features (such as change rate and periodic fluctuations), and break through the feature limitations of single-phase or double-phase comparison, thereby improving the recognition ability of complex change patterns.

[0064] According to the division of the training set, the validation set and the test set, a regional change detection model is constructed, specifically including:

[0065] Based on the pre-trained Changeformer model, end-to-end model parameter learning is performed using the training set, so that the model captures the complex nonlinear mapping relationship between the two periods of remote sensing images related to "change" in the training samples.

[0066] During model training, model selection and hyperparameter optimization are performed based on the validation set, including learning rate (Learning rate), optimizer (Optimizer) and training rounds (Epochs), to prevent overfitting and underfitting and ensure the generalization of the model.

[0067] The trained and optimized model is evaluated for performance using the test set, and the real performance of the model is reflected by evaluation indexes such as IoU (intersection over union), accuracy, precision, recall, F1-score (harmonic mean of precision and recall, which comprehensively measures the precision and recall of the model on different classes, and is used to measure the performance of the model), mAcc (average value of all class Accs; acc refers to the proportion of positive samples compared pixel by pixel), mIoU (average value of all class IoUs), and mF1score (average value of all class F1-scores), and finally a regional change detection model with high precision and high generalization ability is constructed.

[0068] Real-time images of the unmanned aerial vehicle vision are acquired, and a double-time image pair is generated based on the real-time images of the unmanned aerial vehicle vision, specifically including:

[0069] The unmanned aerial vehicle aerial photography platform is docked, a real-time transmission channel of the unmanned aerial vehicle image is established, and aerial photography images of the target monitoring area returned by the unmanned aerial vehicle are received in real time;

[0070] An orthographic image is generated according to the flight route planning of the unmanned aerial vehicle aerial photography task and the real-time aerial photography image;

[0071] The historical images of the target monitoring area are acquired;

[0072] The corresponding regional images are cut out from the historical images according to the orthographic image;

[0073] The cut historical corresponding regional images and the orthographic image are respectively sliced according to a preset training sample image specification to form a double-time image pair.

[0074] Specifically, the orthographic image (DOM) is generated in combination with the flight route planning and the real-time aerial photography image, the geometric precision and spatial reference consistency of the image are ensured, the change detection is carried out based on a unified coordinate system, and errors caused by the viewing angle or the terrain undulation are avoided;

[0075] The corresponding region is cut out from the historical images according to the range of the orthographic image generated in real time, the double-time images are strictly matched in the spatial range, the resolution and the projection system, the pseudo-change interference caused by the regional deviation or the proportion difference is eliminated, the geometric correction (such as terrain undulation compensation) and the radiation correction (such as illumination condition normalization) are implicitly performed in the generation process of the orthographic image, the noise caused by the imaging condition difference (such as season, illumination, sensor parameter) is reduced, and the robustness of the change detection is improved;

[0076] The cut historical image and real-time image are sliced according to preset specifications to generate a standardized double-phase image pair, which meets the input requirements (such as fixed size and multi-scale feature extraction) of the deep learning model, and improves the model training and inference efficiency; by integrating real-time images (reflecting the latest ground state) and historical images (providing baseline data), a time series comparison basis is constructed, enabling the subsequent change detection model to learn the dynamic characteristics (such as gradual or sudden change patterns) of the ground changes, and enhancing the expression ability for complex change types;

[0077] From unmanned aerial vehicle data acquisition, orthographic generation, historical image cutting to slicing processing, full-process automation is realized, manual intervention is reduced, operation cost is reduced, and it is suitable for large-scale and high-frequency monitoring tasks; through orthographic correction and slicing standardization, high-geometric-precision and multi-scale image data are provided, combined with the multi-layer feature extraction capability of the deep learning model, spectral, texture and context semantic features are effectively captured, breaking through the limitations of traditional manual feature design; strict spatial alignment and implicit radiation correction of real-time and historical images reduce false change interference; the automatic process avoids errors introduced by human operation, and improves the detection stability; the end-to-end data processing pipeline from image acquisition to double-phase pair generation does not require human intervention, significantly improves the processing efficiency, and meets the large-scale dynamic monitoring demand.

[0078] In some embodiments, the orthographic image is generated according to the flight route planning of the unmanned aerial vehicle aerial photography task and the real-time aerial photography image, specifically comprising:

[0079] The flight route planning file of the unmanned aerial vehicle aerial photography task is analyzed in real time to obtain the spatial distribution of the flight route, the shooting point position, and the image coverage range information;

[0080] The aerial photography image returned by the unmanned aerial vehicle in real time is subjected to quality inspection, and invalid images with serious influence on imaging quality such as blur, deformation and occlusion are removed to obtain aerial photography images without invalid images;

[0081] According to the shooting point position and the image coverage range in the flight route planning, the aerial photography images without invalid images are automatically grouped, and the images covering adjacent areas or having overlapping parts are grouped into one group;

[0082] The grouped aerial photography images are sequentially subjected to orientation processing to determine the exterior orientation elements of each image; geometric correction is performed to eliminate the geometric distortion of the image; resampling is performed according to a preset ground resolution to unify the scale of the image;

[0083] The images subjected to orientation, correction and resampling processing are spliced according to the grouping to generate an orthographic image covering the target area, and the images can be continuously received and processed during the flight until all the images are returned, forming a complete orthographic image.

[0084] Specifically, by automatically extracting the spatial distribution of the flight path, the shooting point position and the coverage range information, the task parameters are quickly read and analyzed, the manual input errors are reduced, the basic data support is provided for subsequent automatic grouping and splicing, and the automation level of the processing flow is significantly improved; according to the shooting point position and the coverage range, the images are dynamically grouped, the adjacent or overlapping images are grouped and processed, and the calculation resource allocation is optimized; during the flight process, the images are continuously received and processed until the task is completed, the orthographic image is generated in near real time, and the timeliness requirements such as emergency monitoring and rapid response are met;

[0085] By automatically detecting and removing low-quality images such as blur, deformation and occlusion, the noise data interference on subsequent processing is reduced, the signal-to-noise ratio of input data is improved, and the clarity and geometric accuracy of the orthographic image are ensured; the position and attitude parameters of each image are accurately calculated, the image offset caused by the change of the attitude of the unmanned aerial vehicle is eliminated, and the geometric positioning accuracy is improved; according to the preset ground resolution, resampling is performed to eliminate the scale difference caused by multiple time phases or different shooting heights, ensure the consistency of the images in spatial resolution, and provide a standardized data basis for subsequent change detection, feature classification and other applications;

[0086] The shooting point position information in the flight path planning is used to guide the image grouping and splicing, the geometric alignment of adjacent images is ensured, the splicing gap and misalignment phenomenon are reduced, and seamless orthographic images are generated; through geometric correction (such as terrain compensation and lens distortion correction), the image deformation caused by factors such as sensor perspective and terrain undulation is eliminated, the planar accuracy of the orthographic image is improved, and it is more consistent with the real geographical spatial distribution;

[0087] During the flight process, the images are continuously received and processed until all data are returned, the integrity of the orthographic image covering the target area is ensured, the region blanking caused by data loss is avoided, and the monitoring task of a large range and complex terrain is applicable.

[0088] In some embodiments, the historical images of the target monitoring area are acquired, specifically including:

[0089] According to the geographical features of the target monitoring area and the time span requirements of the historical images, historical image sources meeting the conditions are screened from a remote sensing image database, a geographic information data sharing platform or a related historical image archiving agency;

[0090] The projection system and the coordinate system of the historical images are checked, if they are inconsistent with the preset or currently used projection and coordinate system, projection conversion and coordinate registration processing are performed to make the projection and coordinate system of the historical images consistent with the target system;

[0091] The historical images that have been checked and processed are stored in the local or designated server.

[0092] Specifically, according to the geographical features (such as terrain, coverage type) of the target area and the monitoring time span requirements, the matching images are screened from the multi-source database to avoid invalid data processing and reduce the waste of computing resources. For example, in mountainous area monitoring, images with low cloud cover and high resolution are preferentially selected to ensure that the data quality meets the analysis requirements; data can be directly obtained through a shared platform or an archival institution to reduce the time for manual searching and downloading and improve the data acquisition efficiency, which is particularly suitable for large-scale, long-term sequence monitoring tasks;

[0093] Different historical images may use different projections (such as UTM, Albers) or coordinate systems (such as WGS84, local coordinate system), and through conversion and registration, all images can be aligned under the same spatial reference to avoid false change detection caused by coordinate differences; a unified spatial reference system is the basis for accurate registration of double-time or multi-time images, providing reliable geometric protection for subsequent change polygon extraction;

[0094] Different sources of historical images have differences in coordinate systems, resolutions, and coverage ranges, and traditional methods often ignore projection conversion, resulting in detection result deviation. This scheme ensures the spatial consistency of multi-source data through systematic correction, improving the reliability of change detection; automated inspection and conversion process replaces manual registration to avoid positioning deviation caused by subjective operation, which is particularly suitable for large-scale areas or high-frequency monitoring scenarios.

[0095] Based on the regional change detection model, a pair of double-time image pictures is processed to identify the areas where the ground surface has changed, extract the change polygons, and obtain the regional change detection results, which specifically include:

[0096] The pair of double-time image pictures is input into the regional change detection model, which first extracts multi-scale features of the double-time images to obtain spectral, texture, and structural features at different levels; then, by comparing and analyzing the corresponding features of the different time images, the preset threshold or classification rule is used to determine whether the ground surface has changed;

[0097] For the areas determined to have changed, the model uses edge detection and region growing algorithms to accurately outline the boundaries and extract the change polygons;

[0098] According to the image features and the preset change type classification standard, the change type (such as new building, reduced vegetation, etc.) of each change polygon is determined, and the area of each change polygon is calculated;

[0099] The area of each change polygon is calculated, and the change polygon and its related attribute information (including change type, area, location coordinates, etc.) are output in the form of a vector layer, and a statistical report containing detailed change information is generated, thereby obtaining comprehensive and accurate regional change detection results.

[0100] Specifically, the area change detection model automatically extracts spectral, texture and context features in the dual-time images through deep learning or machine learning algorithms, without the need for manual threshold setting or interpretation rules, reduces subjective errors, and realizes full-process automation from image input to change identification; the model directly processes dual-time image pairs, replacing traditional multi-step manual operations (such as image registration, difference calculation, threshold segmentation), significantly shortening processing time, and being suitable for large-scale, high-frequency monitoring scenarios (such as disaster emergency response);

[0101] Through the edge detection or image segmentation algorithm (such as U-Net, Mask R-CNN) built in the model, the change area is pixel-level boundary outlined, and a geometrically regular change polygon is generated, avoiding the boundary blur or fragmentation problem caused by improper threshold selection in traditional methods; the model learns the real change pattern (such as building expansion, vegetation damage) through multi-time comparison, effectively distinguishes real ground changes from sensor noise, seasonal fluctuations and other pseudo changes, and improves the reliability of the detection results;

[0102] The model automatically identifies the change type (such as new building, farmland conversion, water body change) according to the image features (such as spectral response, shape regularity) and the preset classification standard, provides semantic information for subsequent decision-making, and replaces the cumbersome process of manual visual interpretation; based on high-precision change polygons, the geographic spatial area of each polygon is automatically calculated, supporting statistical analysis and compliance verification (such as urban expansion rate monitoring, farmland protection red line evaluation).

[0103] An area change detection system based on unmanned aerial vehicle vision, comprising:

[0104] A data acquisition module for selecting multiple different time remote sensing images of the same monitoring area;

[0105] A data division module for dividing a training set, a verification set and a test set based on the multiple different time remote sensing images;

[0106] A model construction module for constructing an area change detection model according to the division of the training set, the verification set and the test set;

[0107] A real-time image processing module for acquiring real-time images of unmanned aerial vehicle vision, and generating a pair of dual-time image pictures based on the real-time images of unmanned aerial vehicle vision;

[0108] A change detection module for processing the pair of dual-time image pictures based on the area change detection model, identifying the area of the ground surface in the shooting area that has changed, extracting a change polygon, and obtaining an area change detection result.

[0109] A kind of computing device, including memory, processor and computer program stored in the memory and can be run on the processor, when the processor executes the computer program, the steps of any one of the above-mentioned UAV vision-based area change detection method are implemented.

[0110] A computer readable storage medium, the storage medium is stored with computer program, when the computer program is executed by processor, the steps of any one of the above-mentioned UAV vision-based area change detection method are implemented.

[0111] The above only is the preferred embodiment of the present application, and does not use to limit the present application, any modification, equivalent replacement, improvement etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for detecting area change based on unmanned aerial vehicle vision, characterized in that, Specifically comprising: Selecting multiple remote sensing images of different time phases of the same monitoring area, and dividing a training set, a verification set and a test set based on the multiple remote sensing images of different time phases; According to the division of the training set, the verification set and the test set, a regional change detection model is constructed; Obtaining real-time images of unmanned aerial vehicle vision, and generating a pair of double-time-phase image pictures based on the real-time images of unmanned aerial vehicle vision; Processing the pair of double-time-phase image pictures based on the regional change detection model, identifying the region of the ground surface where changes have occurred, extracting change polygons, and obtaining regional change detection results. 2.The UAV vision-based region change detection method of claim 1, wherein, The division of the training set, the verification set and the test set based on the multiple remote sensing images of different time phases specifically comprises: Performing coordinate system and projection unification processing on the multiple remote sensing images of different time phases to obtain unification-processed remote sensing images; Performing slicing operation on the unification-processed remote sensing images and removing invalid data to form an initial multi-time-phase data set; Using a labeling tool suitable for remote sensing double-time-phase change detection to label the double-time-phase data in the initial multi-time-phase data set to obtain a labeled data set; According to a preset proportion relationship, the labeled data set is divided into a training set, a verification set and a test set, wherein the training set is used for parameter learning of the model, the verification set is used for model selection and hyperparameter optimization, and the test set is used for evaluating the generalization ability of the model. 3.The UAV vision-based region change detection method of claim 2, wherein, The construction of the regional change detection model according to the division of the training set, the verification set and the test set specifically comprises: In the pre-trained Changeformer model, the training set is used for end-to-end model parameter learning, so that the Changeformer model captures the nonlinear mapping relationship between the remote sensing images in the training set related to changes; In the model training process, the verification set is used for model selection and hyperparameter optimization, and the hyperparameters include learning rate, optimizer and training rounds; The performance of the trained and optimized model is evaluated using the test set, and the model that passes the performance evaluation is the regional change detection model. 4.The UAV vision-based region change detection method of claim 3, wherein, The generation of a pair of double-time-phase image pictures based on the real-time images of unmanned aerial vehicle vision specifically comprises: Connecting an unmanned aerial vehicle aerial photography platform, establishing a real-time transmission channel for unmanned aerial vehicle images, and real-time receiving aerial photography images returned by the unmanned aerial vehicle for the target monitoring area; Generating an orthographic image according to the flight route planning of the unmanned aerial vehicle aerial photography task and the real-time aerial photography image; Obtaining historical images of the target monitoring area; According to the orthographic image, the corresponding regional image is cut out from the historical image; The cut historical corresponding regional image and the orthographic image are respectively sliced according to a preset training sample picture specification to form a pair of double-time-phase image pictures.

5. The UAV vision-based region change detection method of claim 4, wherein, The generation of an orthographic image according to the flight route planning of the unmanned aerial vehicle aerial photography task and the real-time aerial photography image specifically comprises: Real-time analyzing the flight route planning file of the unmanned aerial vehicle aerial photography task to obtain the spatial distribution, shooting point and image coverage range information of the flight route; Performing quality inspection on the aerial photography images returned by the unmanned aerial vehicle in real time to obtain aerial photography images with invalid images removed; According to the shooting point position and image coverage range in the route planning, the aerial images after removing invalid images are automatically grouped, and images covering adjacent areas or having overlapping parts are grouped into one group; The grouped aerial images are sequentially subjected to orientation, correction and resampling processing to obtain images subjected to orientation, correction and resampling processing; The images subjected to orientation, correction and resampling processing are spliced according to the grouping to generate an orthographic image covering the target area. 6.The UAV vision-based region change detection method of claim 5, wherein, The historical images of the target monitoring area are obtained, specifically including: According to the geographical features of the target monitoring area and the time span requirement of the historical images, historical image sources meeting the conditions are screened from a remote sensing image database, a geographic information data sharing platform or a related historical image archiving agency; The projection system and coordinate system of the historical images are checked, and if they are inconsistent with the preset or currently used projection and coordinate system, projection conversion and coordinate registration processing are performed; The checked and processed historical images are stored in a local or designated server.

7. The UAV vision-based region change detection method of claim 6, wherein, The pair of double-time-phase image pictures is processed based on the regional change detection model to identify the region of the ground surface where changes have occurred, extract the change patches, and obtain the regional change detection result, specifically including: The pair of double-time-phase image pictures is input into the regional change detection model, and the regional change detection model judges whether the ground surface has changed based on the pair of double-time-phase image pictures; The boundaries of the region determined to have changed are outlined, and the change patches are extracted; According to the image features and the preset change type classification standard, the change type to which each change patch belongs is determined, and the area of each change patch is calculated; The change patches and their related attribute information are output in the form of a vector layer.

8. A drone vision-based region change detection system, comprising: It includes: a data acquisition module for selecting multiple remote sensing images of different time phases of the same monitoring area; a data division module for dividing a training set, a verification set and a test set based on the multiple remote sensing images of different time phases; a model construction module for constructing a regional change detection model according to the division of the training set, the verification set and the test set; a real-time image processing module for acquiring real-time images of unmanned aerial vision and generating a pair of double-time-phase image pictures based on the real-time images of unmanned aerial vision; a change detection module for processing the pair of double-time-phase image pictures based on the regional change detection model to identify the region of the ground surface where changes have occurred, extract the change patches, and obtain the regional change detection result.

9. A computing device, comprising: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to realize the steps of the regional change detection method based on unmanned aerial vision in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is executed by the processor to realize the steps of the regional change detection method based on unmanned aerial vision in any one of claims 1-7.

Images