Land inspection method

Abstract

The invention discloses a land inspection method. The method comprises the steps of performing image shooting on a key monitoring area by using an unmanned aerial vehicle to form an image data set; performing image shooting on the inspection area by using the unmanned aerial vehicle, and then inputting the deep learning model for training to form an unmanned aerial vehicle land inspection model; and preprocessing a to-be-detected land inspection image shot by the unmanned aerial vehicle, and inputting the preprocessed to-be-detected land inspection image into the unmanned aerial vehicle land inspection model for identification. The method has the advantages that the accuracy of a land inspection analysis algorithm can be improved, and the accuracy rate can reach 95%. The method can improve the inspection efficiency of the land, remarkably improves the inspection efficiency of the ownership land of the oil region, improves the inspection area, and reduces the problem that the personnel cannot reach the normal inspection due to environmental factors. Through a land inspection analysis algorithm, targets can be quickly positioned, problems can be timely found, the post-processing cost is effectively reduced, and the targets of cost reduction, throttling and income creation are achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of land inspection and monitoring, and in particular relates to a land inspection method. Background Art

[0002] With the continuous improvement of the refinement and informatization of land resource management, the frequency of collection and update of land-related information data is also required to be higher. The traditional ground survey and inspection methods are costly and inefficient, and can no longer meet the current land management needs. At present, satellite remote sensing technology has entered the stage of business operation in land inspection and law enforcement, such as the "National Land Change Survey and Remote Sensing Monitoring Work" carried out by the Ministry of Land and Resources every year. This work has the characteristics of long cycle and fixed time. Satellite remote sensing has become the mainstream technology in land information survey. UAV inspection has the ability to obtain data over a large range, objectively and repeatedly, and can quickly obtain high-resolution remote sensing images to replace manual or satellite remote sensing inspections. After shooting land images, the remote sensing images are analyzed to determine which areas of land have illegal acts.

[0003] The prior art discloses a drone inspection method and inspection system that can be used for land inspection. The system realizes automatic inspection of drones along the inspection route, and automatically stops at stations according to the inspection route to replace batteries. It does not require human control, nor does it need to set up management personnel at each station, saving a lot of manpower and material resources, and saving drone charging time, thereby shortening inspection time and improving inspection efficiency. However, it is only limited to strengthening the automated battery replacement of drone inspections and saving drone charging time. It is an improvement in drone power modules and ground charging stations, but there is no improvement in the image recognition model level. It still uses manual interpretation, which has the problem of manual missed judgments and wrong judgments, and fails to use image recognition technology to improve the level of intelligent recognition.

[0004] In addition, the prior art also discloses a large-scale illegal target identification and processing method, which includes obtaining an image within a preset range within a preset period, the image includes multiple sub-images, and each sub-image includes at least a tower pole image; based on the image recognition model, determining whether there are suspicious objects in the multiple sub-images, and the suspicious objects include buildings and construction tools; if there are suspicious objects in at least two sub-images, then determining whether there are repeated suspicious objects in all suspicious objects; if there are repeated suspicious objects, then performing deduplication processing on the repeated suspicious objects, retaining one repeated suspicious object, and determining the geographical location of the retained repeated suspicious object and the non-repeated suspicious object based on the geographical location of the tower pole in the image as the illegal target location. This method can improve the efficiency and timeliness of illegal construction monitoring, but can only judge the suspicious elements in multiple images of the same position sequence within the set period, and can only judge the presence of suspicious objects at the position, and cannot identify the type of suspicious objects through key points, and the splicing processing efficiency is poor, and no accurate object-oriented abnormal marking area recognition algorithm is used, so it fails to fundamentally solve the technical problems of image key point extraction and abnormal recognition of land inspection.

[0005] Chinese patent application No. 202211588488.5 discloses a "water conservancy project safety monitoring method and platform based on data processing", which includes: generating corresponding plane images according to the land area and river area corresponding to the water conservancy project, and determining the land position points of the land area and the river position points of the river area respectively; generating the river inspection path of the unmanned inspection ship and the land inspection path of the inspection robot based on the first inspection list; determining the target inspection position of the target inspection equipment according to the type of coverage point area, and the target inspection equipment is an unmanned inspection ship and / or an inspection robot; if the first target number and the second target number do not correspond, output the first monitoring information and the corresponding second land inspection position point or the second river inspection position point; if the server determines that there is a dynamic target in the second image data collected by the unmanned inspection ship or the inspection robot, output the second monitoring information and the corresponding first land inspection position point or the first river inspection position point. The invention needs to collect a large amount of data, including indicators such as water level, water flow, and rainfall. If the data collection is not comprehensive or the data quality is inaccurate, it will lead to errors in the monitoring results. Processing and analyzing the collected data is an important part of ensuring monitoring accuracy, but the current data processing algorithms are not precise enough to meet actual monitoring needs. Summary of the invention

[0006] In order to solve the above problems, the object of the present invention is to provide a land inspection method.

[0007] In order to achieve the above object, the land inspection method provided by the present invention comprises the following steps performed in sequence:

[0008] 1) Use drones to capture images of key monitoring areas where abnormal land behaviors, including earthwork, soil dumping, illegal construction, river garbage, and illegal water storage in fish ponds, are found within the inspection area, and then process the images to form an image data set;

[0009] 2) Use drones to capture images of the inspection area, then pre-process the images and input them into a deep learning model for training to form a drone land inspection model;

[0010] 3) The land inspection images to be detected taken by the drone are pre-processed and input into the drone land inspection model, and then the single target detection method is used to extract the key points of the pre-processed land inspection images to be detected and the land key monitoring images pre-input into the drone land inspection model. After that, the multi-region matching mechanism technology is used to match the key points to obtain the common areas, and the common areas are divided into common blocks. The abnormal position monitoring technology is then used to analyze the common block images, and finally the abnormal land area and position are obtained and an alarm is issued.

[0011] In step 1), the drone is used to capture images of key monitoring areas in the inspection area where abnormal land behaviors such as earthwork, soil dumping, illegal construction, river garbage, and illegal water storage in fish ponds exist, and then the images are processed to form an image data set. The specific steps are as follows:

[0012] 1.1) Determine the key monitoring areas that need to be photographed through on-site surveys, map analysis, etc., and determine the purpose of shooting based on actual needs;

[0013] 1.2) Select appropriate drones and camera equipment according to the key monitoring areas and purposes;

[0014] 1.3) Plan routes based on the actual conditions of key monitoring areas;

[0015] 1.4) Control the drone to fly along the above route, and use the camera equipment it carries to take aerial photos of key monitoring areas where abnormal land activities, including earthworks, dumping of debris, illegal construction, river garbage, and illegal water storage in fish ponds, are found, to obtain original key land monitoring images;

[0016] 1.5) The above original land key monitoring images are processed including denoising, color correction and alignment to improve the image quality and consistency;

[0017] 1.6) The locations and features of abnormal land behaviors in the processed images are marked and then divided into different categories for the convenience of subsequent image processing and analysis. The image dataset is composed of all the marked and classified key land monitoring images.

[0018] In step 2), the drone is used to take images of the inspection area, and then the images are preprocessed and then input into the deep learning model for training. The specific steps of forming the drone land inspection model are as follows:

[0019] 2.1) Control the drone to fly along the inspection area, and use the camera equipment carried to take aerial photos of the inspection area to obtain the original land inspection images;

[0020] 2.2) Marking the locations of land bumps, illegal buildings and other environmental changes on the above original land inspection images; the marking is done manually or with the help of some automatic marking tools;

[0021] 2.3) The annotated original land inspection images are preprocessed including scaling, cropping, and normalization to facilitate subsequent model training. All preprocessed land inspection images constitute a land inspection image dataset;

[0022] 2.4) Input the above land inspection image dataset into the deep learning model YOLO v5 for training, and adjust the model parameters appropriately according to actual needs to obtain the original drone land inspection model;

[0023] 2.5) Evaluate the original UAV land inspection model mentioned above. The evaluation indicators include accuracy, recall rate, and precision. Use the evaluation results to adjust the model parameters to optimize the model structure and obtain the UAV land inspection model.

[0024] In step 3), the land inspection image to be detected taken by the drone is preprocessed and input into the drone land inspection model, and then the single target detection method is used to extract the key points of the preprocessed land inspection image to be detected and the land key monitoring image pre-input into the drone land inspection model, and then the multi-region matching mechanism technology is used to match the key points to obtain the common area, and the common area is divided into common blocks, and then the abnormal position monitoring technology is used to analyze the common block image, and finally the specific steps of obtaining the abnormal land area and position and alarming are as follows:

[0025] 3.1) Control the drone to conduct routine inspections in the inspection area, and use the camera equipment carried to take real-time aerial photos to obtain inspection images of the land to be inspected;

[0026] 3.2) The above-mentioned land inspection images to be detected are pre-processed including scaling, cropping, and normalization, and then input into the UAV land inspection model;

[0027] 3.3) The UNet model in the single target detection method is used to extract the key points of the pre-processed land inspection image to be detected and the key land monitoring image pre-input into the UAV land inspection model;

[0028] 3.4) Using the RANSAC algorithm in the multi-region matching mechanism technology to match the above key points, obtain the common region, and divide the common region image into multiple common block images;

[0029] 3.5) The PatchCore algorithm in the abnormal location monitoring technology is used to analyze the common block image, and finally the abnormal land area and location are obtained and an alarm is issued.

[0030] The land inspection method provided by the present invention has the following beneficial effects:

[0031] (1) It can improve the accuracy of land inspection analysis algorithms. In terms of land inspection target detection, the algorithm uses a combination of single-stage target detection technology, key point extraction technology, multi-region matching mechanism technology, and abnormal position monitoring technology to greatly improve the accuracy of identifying targets such as illegal land construction and illegal activities, with an accuracy rate of up to 95%.

[0032] (2) It can improve the efficiency of land inspection and supervision. By using drones to conduct intelligent land inspections, compared with manual inspections, it can significantly improve the inspection efficiency of oilfield-owned land, increase the inspection area, and reduce the problem of personnel being unable to reach the site and unable to conduct normal inspections due to environmental factors.

[0033] (3) Through the land inspection and analysis algorithm, the target can be quickly located, problems can be discovered in a timely manner, and the cost of post-processing can be effectively reduced, thus achieving the goals of cost reduction, savings and revenue generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flow chart of the land inspection method provided by the present invention. DETAILED DESCRIPTION

[0035] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] like Figure 1 As shown, the land inspection method provided by the present invention comprises the following steps performed in sequence:

[0037] 1) Use drones to capture images of key monitoring areas where abnormal land behaviors, including earthwork, soil dumping, illegal construction, river garbage, and illegal water storage in fish ponds, are found within the inspection area, and then process the images to form an image data set;

[0038] The specific steps are as follows:

[0039] 1.1) Determine the key monitoring areas that need to be photographed through on-site surveys, map analysis, etc., and determine the purpose of shooting based on actual needs;

[0040] 1.2) Select appropriate drones and video equipment according to the key monitoring areas and purposes; the selection of drones should take into account factors such as flight stability, maneuverability, and operation time, and the selection of video equipment should take into account factors such as resolution, pixels, and viewing angle;

[0041] 1.3) Plan routes based on the actual conditions of key monitoring areas; the planned routes need to take into account factors such as aerial photography altitude, flight speed, and shooting angle to ensure that sufficient quantity and quality of images are captured;

[0042] 1.4) Control the drone to fly along the above route, and use the camera equipment it carries to take aerial photos of key monitoring areas where abnormal land activities, including earthworks, dumping of debris, illegal construction, river garbage, and illegal water storage in fish ponds, are found, to obtain original key land monitoring images;

[0043] 1.5) The above original land key monitoring images are processed including denoising, color correction and alignment to improve the image quality and consistency;

[0044] 1.6) The locations and features of abnormal land behaviors in the processed images are marked and then divided into different categories for the convenience of subsequent image processing and analysis. The image dataset is composed of all marked and classified key land monitoring images; the image dataset should include a sufficient number of key land monitoring images of different categories to meet actual needs.

[0045] 2) Use drones to capture images of the inspection area, then pre-process the images and input them into a deep learning model for training to form a drone land inspection model;

[0046] The specific steps are as follows:

[0047] 2.1) Control the drone to fly along the inspection area, and use the camera equipment carried to take aerial photos of the inspection area to obtain the original land inspection images;

[0048] 2.2) Marking the locations of land bumps, illegal buildings and other environmental changes on the above original land inspection images; the marking is done manually or with the help of some automatic marking tools;

[0049] 2.3) The annotated original land inspection images are preprocessed including scaling, cropping, and normalization to facilitate subsequent model training. All preprocessed land inspection images constitute a land inspection image dataset;

[0050] 2.4) Input the above land inspection image dataset into the deep learning model YOLO v5 for training. During the training process, attention should be paid to model parameter setting, overfitting and underfitting problems, and the model parameters should be appropriately adjusted according to actual needs to obtain the original UAV land inspection model;

[0051] 2.5) Evaluate the original UAV land inspection model, with the evaluation indicators including accuracy, recall rate, and precision rate, and use the evaluation results to adjust the model parameters to optimize the model structure and obtain the UAV land inspection model;

[0052] 3) The land inspection images to be detected taken by the drone are pre-processed and input into the drone land inspection model, and then the single target detection method is used to extract the key points of the pre-processed land inspection images to be detected and the key land monitoring images pre-input into the drone land inspection model. After that, the multi-region matching mechanism technology is used to match the key points to obtain the common areas, and the common areas are divided into common blocks. Then, the abnormal position monitoring technology is used to analyze the common block images, and finally the abnormal land area and position are obtained and an alarm is issued;

[0053] The specific steps are as follows:

[0054] 3.1) Control the drone to conduct routine inspections in the inspection area, and use the camera equipment carried to take real-time aerial photos to obtain inspection images of the land to be inspected;

[0055] 3.2) The above-mentioned land inspection images to be detected are pre-processed including scaling, cropping, and normalization, and then input into the UAV land inspection model;

[0056] 3.3) The UNet model in the single target detection method is used to extract the key points of the pre-processed land inspection image to be detected and the key land monitoring image pre-input into the UAV land inspection model;

[0057] 3.4) Using the RANSAC algorithm in the multi-region matching mechanism technology to match the above key points, obtain the common region, and divide the common region image into multiple common block images;

[0058] 3.5) The PatchCore algorithm in the abnormal location monitoring technology is used to analyze the common block image, and finally the abnormal land area and location are obtained and an alarm is issued.

Claims

1. A land inspection method, characterized in that: The land inspection method comprises the following steps performed in sequence: 1) Use drones to capture images of key monitoring areas where abnormal land behaviors, including earthwork, soil dumping, illegal construction, river garbage, and illegal water storage in fish ponds, are found within the inspection area, and then process the images to form an image data set; 2) Use drones to capture images of the inspection area, then pre-process the images and input them into a deep learning model for training to form a drone land inspection model; 3) The land inspection images to be detected taken by the drone are pre-processed and input into the drone land inspection model, and then the single target detection method is used to extract the key points of the pre-processed land inspection images to be detected and the land key monitoring images pre-input into the drone land inspection model. After that, the multi-region matching mechanism technology is used to match the key points to obtain the common areas, and the common areas are divided into common blocks. The abnormal position monitoring technology is then used to analyze the common block images, and finally the abnormal land area and position are obtained and an alarm is issued.

2. The land inspection method according to claim 1, characterized in that: In step 1), the drone is used to capture images of key monitoring areas in the inspection area where abnormal land behaviors such as earthwork, soil dumping, illegal construction, river garbage, and illegal water storage in fish ponds exist, and then the images are processed to form an image data set. The specific steps are as follows: 1.1) Determine the key monitoring areas that need to be photographed through on-site surveys, map analysis, etc., and determine the purpose of shooting based on actual needs; 1.2) Select appropriate drones and camera equipment according to the key monitoring areas and purposes; 1.3) Plan routes based on the actual conditions of key monitoring areas; 1.4) Control the drone to fly along the above route, and use the camera equipment it carries to take aerial photos of key monitoring areas where abnormal land activities, including earthworks, dumping of debris, illegal construction, river garbage, and illegal water storage in fish ponds, are found, to obtain original key land monitoring images; 1.5) The above original land key monitoring images are processed including denoising, color correction and alignment to improve the image quality and consistency; 1.6) The locations and features of abnormal land behaviors in the processed images are marked and then divided into different categories for the convenience of subsequent image processing and analysis. The image dataset is composed of all the marked and classified key land monitoring images.

3. The land inspection method according to claim 1, characterized in that: In step 2), the drone is used to take images of the inspection area, and then the images are preprocessed and then input into the deep learning model for training. The specific steps of forming the drone land inspection model are as follows: 2.1) Control the drone to fly along the inspection area, and use the camera equipment carried to take aerial photos of the inspection area to obtain the original land inspection images; 2.2) Marking the locations of land bumps, illegal buildings and other environmental changes on the above original land inspection images; the marking is done manually or with the help of some automatic marking tools; 2.3) The annotated original land inspection images are preprocessed including scaling, cropping, and normalization to facilitate subsequent model training. All preprocessed land inspection images constitute a land inspection image dataset; 2.4) Input the above land inspection image dataset into the deep learning model YOLO v5 for training, and adjust the model parameters appropriately according to actual needs to obtain the original drone land inspection model; 2.5) Evaluate the original UAV land inspection model mentioned above. The evaluation indicators include accuracy, recall rate, and precision. Use the evaluation results to adjust the model parameters to optimize the model structure and obtain the UAV land inspection model.

4. The land inspection method according to claim 1, characterized in that: In step 3), the land inspection image to be detected taken by the drone is preprocessed and input into the drone land inspection model, and then the single target detection method is used to extract the key points of the preprocessed land inspection image to be detected and the land key monitoring image pre-input into the drone land inspection model, and then the multi-region matching mechanism technology is used to match the key points to obtain the common area, and the common area is divided into common blocks, and then the abnormal position monitoring technology is used to analyze the common block image, and finally the specific steps of obtaining the abnormal land area and position and alarming are as follows: 3.1) Control the drone to conduct routine inspections in the inspection area, and use the camera equipment carried to take real-time aerial photos to obtain inspection images of the land to be inspected; 3.2) The above-mentioned land inspection images to be detected are pre-processed including scaling, cropping, and normalization, and then input into the UAV land inspection model; 3.3) The UNet model in the single target detection method is used to extract the key points of the pre-processed land inspection image to be detected and the key land monitoring image pre-input into the UAV land inspection model; 3.4) Using the RANSAC algorithm in the multi-region matching mechanism technology to match the above key points, obtain the common region, and divide the common region image into multiple common block images; 3.5) The PatchCore algorithm in the abnormal location monitoring technology is used to analyze the common block image, and finally the abnormal land area and location are obtained and an alarm is issued.

Images