Agricultural product storage method and system based on unmanned aerial vehicle safety management

Abstract

The invention relates to the technical field of image recognition, in particular to an agricultural product storage method and system based on unmanned aerial vehicle safety management, and the method comprises the steps: dividing a target region into a plurality of sub-regions, endowing each sub-region with a unique region ID and a first cruise path, and setting a second cruise path for the target region; the unmanned aerial vehicle obtains the mode of the patrol task after being started; if the patrol task is in the first mode, the unmanned aerial vehicle goes to the corresponding sub-areas in sequence according to the area ID included in the patrol task, and shoots a first environment video according to the corresponding first cruise path; if the patrol task is in a second mode, the unmanned aerial vehicle performs high-angle shooting on the target area based on a second cruise path to obtain a second environment video; and analyzing the first environment video and the second environment video to respectively obtain a first analysis result and a second analysis result. According to the invention, the internal environment and the external environment of the grain depot can be quickly patrolled, so that the safety management level of the grain depot is improved.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and more specifically to an agricultural product storage system based on drone safety management. Background Art

[0002] With the promotion of modern agricultural technology, the output of agricultural products has shown a steady growth trend. However, the production of agricultural products is highly seasonal. A large number of products are often concentrated in the harvest season, but consumer demand is relatively evenly distributed throughout the year. This leads to the need for proper storage of agricultural products after harvest, so grain depots need to be monitored.

[0003] For example, the Chinese invention patent with publication number CN115407800A discloses a drone inspection system and inspection method in an agricultural product storage and preservation warehouse, including an image information acquisition module, a model training module, a real-time detection module, a data processing module, a terminal display module and a drone alarm module. The image information acquisition module is connected to the real-time detection module, and the real-time detection module is respectively connected to the model training module and the data processing module, and the data processing module is connected to the terminal display module, and the terminal display module is connected to the drone alarm module. The above scheme effectively solves the problems of high work risk and high work intensity in manual inspection, and can achieve all-weather and all-time inspection.

[0004] At present, grain is generally transported to grain depots by grain trucks. Since grain trucks will receive grain transportation subsidies when they enter grain depots for the first time, some grain trucks have been prompted to change their license plates in order to enter grain depots multiple times to defraud subsidies, which will eventually cause economic damage to the grain depots. Summary of the invention

[0005] In view of the above problems, the present invention provides an agricultural product storage method and system based on drone safety management to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: A method for storing agricultural products based on drone safety management, comprising: Set a target area, divide the target area into multiple sub-areas, record the coverage of each sub-area, and assign a unique area ID to each sub-area; A first cruising path is set for each of the sub-areas, and a second cruising path is set for the target area; The patrol mission is divided into the first mode and the second mode, and the mode of the patrol mission is obtained after the drone is started; If the patrol mission is the first mode, the drone goes to the corresponding sub-areas in sequence according to the area ID included in this patrol mission, obtains the corresponding first cruise path according to the area ID, and shoots the first environment video within the coverage of the sub-area based on the first cruise path; If the patrol mission is the second mode, the drone takes a bird's-eye view of the target area based on the second cruise path to obtain a second environment video; The first environmental video and the second environmental video are analyzed to obtain a first analysis result and a second analysis result respectively, wherein the first analysis result includes the situation of the grain transport vehicle in the grain warehouse, and the second analysis result includes the integrity of the grain warehouse building.

[0007] Further, obtaining the first analysis result includes the following steps: Splitting the first environment video into a plurality of first images based on a preset time interval, identifying the first images, and acquiring the grain transport vehicle included therein; A second image containing only a single grain truck is cropped from the first image, and the grain truck in each of the second images is analyzed to determine a suspicious object therein, and an area where the suspicious object is located is defined as a suspicious area. The drone goes to the suspicious area, surrounds the suspicious object, and takes high-resolution images from multiple angles; The high-resolution image is analyzed to authenticate the identity of the suspicious object. If the identity authentication fails, a warning message is generated and sent to a preset terminal.

[0008] Furthermore, analyzing the grain transport vehicle includes the following steps: When the grain transport vehicle enters the warehouse, a third image is taken from multiple angles, the third image is stored and a monitoring time is set; After acquiring the second image, the drone acquires the third image which is still within the monitoring time, extracts a bird's-eye view from the third image and defines it as a historical image, calculates a first similarity between the second image and each of the historical images, and marks the grain truck as the suspicious object if the first similarity is greater than a first threshold.

[0009] Furthermore, analyzing the high-resolution image comprises the following steps: Assigning a first number to each of the high-resolution images, assigning a second number to each of the third images, extracting the first high-resolution image with the first number, and calculating the second similarities with each of the third images with the second numbers, and extracting the second similarity with the largest value as the third similarity corresponding to the first high-resolution image with the first number; Extract the second high-resolution image of the first number, calculate the second similarity with each of the third images of the second number, and determine the third similarity therefrom, repeat this step until all the high-resolution images are extracted, calculate the average value of the third similarities, and if the average value is greater than a second threshold value and the license plate number of the suspicious object appears for the first time, determine that the identity authentication of the suspicious object has failed.

[0010] Further, analyzing the second environment video includes the following steps: A first standard image of the grain storage building is captured in advance, the second environment video is split into multiple fourth images based on the preset time interval, the first standard image and the fourth image are gray-scaled to obtain a second standard image and a fifth image, the fifth image is segmented into multiple building outlines based on a contour segmentation algorithm, a sixth image containing only one building outline is captured from the fifth image, the sixth image and the second standard image are processed to the same size and subtracted to obtain a difference image, and if the difference image has multiple consecutive pixel points whose pixel points are greater than a third threshold, it is determined that the grain storage building is abnormal.

[0011] Furthermore, segmenting the outline building comprises the following steps: The edge pixel points in the fifth image are determined, and the edge pixel points that are adjacent in position are sequentially connected to complete the segmentation of the fifth image.

[0012] Furthermore, determining the edge pixel point comprises the following steps: If there is a target pixel point, and the neighboring pixel points around the target pixel point meet a preset condition, the target pixel point is defined as the edge pixel point, and the preset condition is that the difference in pixel values ​​of the neighboring pixel points is greater than a fourth threshold.

[0013] Furthermore, identifying the first image includes the following steps: A picture set containing the grain truck is obtained, and a training data set is obtained after annotation. A YOLO network model is established, the constructed YOLO network model is trained, and the checkpoint file generated during the training process is tested. Finally, the checkpoint file with the highest mAP is selected as the final network model, and the first image is recognized based on the final network model.

[0014] Furthermore, the final network model includes a backbone network, a Neck, and a prediction output layer. The backbone network adopts the MobileNetv3 network structure. The Neck includes three side output layers and an output layer. The side output layers are the SPP layer, the feature pyramid layer, and the upsampling layer.

[0015] The present invention also provides an agricultural product storage system based on drone safety management, which is used to implement the above-mentioned agricultural product storage method based on drone safety management, and the system includes: A setting module, used to set a target area, divide the target area into multiple sub-areas, record the coverage of each sub-area, assign a unique area ID to each sub-area, set a first cruising path for each sub-area, and set a second cruising path for the target area; The patrol module divides the patrol mission into a first mode and a second mode. After the UAV is started, the mode of the patrol mission is obtained. If the patrol mission is the first mode, the UAV goes to the corresponding sub-areas in sequence according to the area ID included in the patrol mission, and obtains the corresponding first cruise path according to the area ID, and shoots the first environment video within the coverage of the sub-area based on the first cruise path. If the patrol mission is the second mode, the UAV takes a bird's-eye view of the target area based on the second cruise path to obtain the second environment video; The analysis module is used to analyze the first environmental video and the second environmental video to obtain a first analysis result and a second analysis result respectively. The first analysis result includes the situation of the grain transport vehicle in the grain warehouse, and the second analysis result includes the integrity of the grain warehouse building.

[0016] Compared with the prior art, the beneficial effects of the present invention are at least as follows: Before using the drone for patrol, the present invention enables the drone to automatically perform patrol tasks through preset cruise paths and mission modes, thereby reducing the need for manual operation and improving efficiency. By dividing the target area into multiple sub-areas and setting a unique ID and cruise path for each sub-area, it is ensured that the drone can cover every corner of the grain depot. During patrol, the drone can shoot environmental videos in real time. By analyzing the environmental videos, the behavior of grain transport vehicles in the grain depot can be monitored to prevent the occurrence of suspicious behaviors. It can also check the integrity of the grain depot building to ensure the safety of the grain storage environment.

[0017] The present invention can quickly reach the designated area for shooting and analysis through the drone, thereby improving the response speed to emergencies. The drone can work under different weather and lighting conditions, thereby improving the adaptability and reliability of the patrol system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flowchart of the steps of an agricultural product storage method based on drone safety management of the present invention; Figure 2 This is a structural schematic diagram of an agricultural product storage system based on drone safety management of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] like Figure 1 As shown, a method for storing agricultural products based on drone safety management includes: S1: Set the target area, divide the target area into multiple sub-areas, record the coverage of each sub-area, and assign a unique area ID to each sub-area.

[0021] The target area in this embodiment is the entire grain storage area, which includes multiple granaries. According to the locations of the granaries, the target area is divided into multiple rectangular sub-areas, and then the coverage range of each sub-area is recorded. Specifically, a coordinate system is established in the target area, and the coverage range of the sub-area is represented by the four vertices of the rectangle.

[0022] S2: Setting a first cruise path for each sub-area and a second cruise path for the target area.

[0023] S3: The patrol mission is divided into a first mode and a second mode, and the UAV is started to obtain the mode of this patrol mission.

[0024] The first cruise path is the cruise path of the drone in the grain depot, and the second cruise path is the path for the drone to cruise the entire target area. The cruise path can be determined according to the building shape of the grain depot. When setting it, it should be followed that the drone can patrol every corner of the grain depot through the cruise path. By letting the drone patrol in the grain depot, you can clearly see every corner of the grain depot, so as to promptly discover and deal with violations. By patrolling the entire target area with a drone, you can check whether there is damage to the roof or wall, so as to ensure the environmental safety of the grain in the grain depot. By setting the first mode and the second mode, when you need to use a drone for patrol, you only need to select the corresponding mode in advance, and the drone can automatically perform subsequent patrol tasks, thereby improving the convenience of drone patrol operations.

[0025] S4: If the patrol mission is the first mode, the UAV goes to the corresponding sub-areas in sequence according to the area ID included in this patrol mission, obtains the corresponding first cruise path according to the area ID, and shoots the first environment video within the coverage of the sub-area based on the first cruise path.

[0026] S5: If the patrol mission is the second mode, the drone takes a bird's-eye view of the target area based on the second cruise path to obtain a second environment video.

[0027] Before patrolling, the operator sets the patrol mission to the first mode and inputs at least one area ID as the target area for this patrol; after the UAV is started, if it is detected that the current patrol mission is the first mode, the area ID of the sub-area to be patrolled is obtained, and the corresponding coverage range is obtained according to the area ID. After reaching the coverage range, the interior is photographed according to the preset first cruise path, thereby obtaining the first environmental video of the grain warehouse.

[0028] If it is the second mode, the drone will increase its altitude until it is above the roof of the grain warehouse, and then take a bird's-eye view of the entire target area according to the preset second cruise path, thereby obtaining a second environmental video of the entire grain warehouse.

[0029] S6: Analyze the first environment video and the second environment video to obtain a first analysis result and a second analysis result respectively. The first analysis result includes the situation of the grain transport vehicle in the grain warehouse, and the second analysis result includes the integrity of the grain warehouse building.

[0030] The specific method of analyzing the environmental video will be introduced later. By analyzing the first environmental video, it can be determined whether there are suspicious behaviors of grain transport vehicles in the grain depot, such as changing license plates and entering the grain depot multiple times. By analyzing the second environmental video, it can be determined whether there is damage on the roof surface of each grain depot.

[0031] Before using the drone for patrol, the present invention enables the drone to automatically perform patrol tasks through preset cruise paths and mission modes, thereby reducing the need for manual operation and improving efficiency. By dividing the target area into multiple sub-areas and setting a unique ID and cruise path for each sub-area, it is ensured that the drone can cover every corner of the grain depot. During patrol, the drone can shoot environmental videos in real time. By analyzing the environmental videos, the behavior of grain transport vehicles in the grain depot can be monitored to prevent the occurrence of suspicious behaviors. It can also check the integrity of the grain depot building to ensure the safety of the grain storage environment.

[0032] The present invention can quickly reach the designated area for shooting and analysis through the drone, thereby improving the response speed to emergencies. The drone can work under different weather and lighting conditions, thereby improving the adaptability and reliability of the patrol system.

[0033] It is particularly noteworthy that the present invention can quickly inspect the internal and external environments of the grain depot, thereby improving the safety management level of the grain depot.

[0034] In this embodiment, obtaining the first analysis result includes the following steps: The first environment video is split into a plurality of first images based on a preset time interval, the first images are identified, and the grain transport trucks included therein are obtained.

[0035] A second image containing only a single grain truck is cropped out from the first image, and the grain trucks in each second image are analyzed to determine the suspicious objects therein. The area where the suspicious objects are located is defined as a suspicious area. The drone goes to the suspicious area, surrounds the suspicious objects and takes high-resolution images from multiple angles.

[0036] In this embodiment, recognizing the first image includes the following steps: Get a set of pictures containing grain trucks, annotate them to get a training data set, build a YOLO network model, train the constructed YOLO network model, test the checkpoint file generated during the training process, and finally select the checkpoint file with the highest mAP as the final network model. Recognize the first image based on the final network model.

[0037] The final network model includes the backbone network, Neck, and prediction output layer. The backbone network adopts the MobileNetv3 network structure. Neck includes three side output layers and one output layer. The side output layers are SPP layer, feature pyramid layer and upsampling layer.

[0038] The preset time interval is, for example, 1s. After the drone records the first environment video, the first environment video is transmitted to the analysis server. The analysis server splits the first environment video into multiple static first images at intervals of 1s, and uses the YOLO neural network algorithm to identify each first image, thereby determining which first images include grain trucks.

[0039] After that, the area including the grain truck is cut out from the first image to obtain a second image including only the grain truck. By cutting out the second image, the interference factors in the first image except for the grain truck to be analyzed can be eliminated, which is convenient for subsequent comparative analysis. By analyzing each grain truck, the suspicious object is determined. The suspicious object in this embodiment is that the grain truck may have entered the grain depot multiple times. After determining that there is a suspicious object in the grain depot, the analysis server sends the suspicious area where the suspicious object is located to the drone. After receiving the data, the drone goes to the surroundings of the corresponding grain truck and performs multi-angle surround shooting. Specifically, the front view, rear view, left view and right view of the grain truck are obtained by circling and shooting at close range.

[0040] High-resolution images are analyzed to authenticate suspicious objects. If the authentication fails, a warning message is generated and sent to the preset terminal.

[0041] Surround shooting can be used to obtain the license plate number and more subtle external image of the suspicious object. During identity authentication, if the license plate number of the suspicious object appears for the first time, but its external image determines that the grain transport truck is not entering the grain depot for the first time, then the identity authentication of the grain transport truck is judged to have failed. There may be a situation where the license plate is replaced to forge the first entry into the grain depot in order to defraud subsidies. Therefore, the license plate and location of the grain transport truck are sent to the remote terminal to remind relevant personnel to check it in time.

[0042] The present embodiment analyzes the grain transport vehicle and includes the following steps: When the grain truck enters the warehouse, a third image is taken from multiple angles, the third image is stored and a monitoring time is set.

[0043] After acquiring the second image, the drone acquires the third image which is still within the monitoring time, extracts the bird's-eye view from the third image and defines it as a historical image, calculates the first similarity between the second image and each historical image, and marks the grain truck as a suspicious object if the first similarity is greater than the first threshold.

[0044] When the grain truck enters the warehouse for the first time, the staff can take a third image of it from multiple angles, or the surveillance camera at the entrance can take a third image from multiple angles and record the license plate number of the grain truck. After obtaining the third image, the third image is sent to the storage server for storage, and a monitoring time is set. The monitoring time can be set according to actual needs. In this embodiment, the monitoring time is set to 6 months. For example, if the third image is taken on January 1, the monitoring time is from January 2 to July 2. Through the above settings, when the grain truck enters the grain warehouse for the first time, even if the drone takes the second image, since the corresponding third image is not within the monitoring time, it will not be misjudged as multiple entries. If the grain truck enters the grain warehouse again after the monitoring time has expired, it will not be judged as multiple entries.

[0045] When making the comparison, since the drone first takes a bird's-eye view inside the grain warehouse, based on this, when comparing the first similarity, the bird's-eye view is first extracted from the third image as the historical image, and then the second image is compared with the first similarity of each historical image that has been stored and within the monitoring time through the convolutional neural network algorithm. The code and specific implementation method of the convolutional neural algorithm for comparing image similarity have been made public on Github, or Matlab software can be used, in which parameters can be input to quickly construct a convolutional neural network; in this embodiment, the first threshold is set to 0.95. When the first similarity between the historical image and the second image is greater than 0.95, the grain truck corresponding to the second image is regarded as a suspicious object.

[0046] In this embodiment, analyzing the high-resolution image includes the following steps: A first number is assigned to each high-resolution image, a second number is assigned to each third image, the first high-resolution image with the first number is extracted, and the second similarity with each second-numbered third image is calculated, and the second similarity with the largest value is extracted as the third similarity corresponding to the first high-resolution image with the first number.

[0047] Extract the second high-resolution image with the first number, and calculate the second similarity with each second-numbered third image, and determine the third similarity therefrom, repeat this step until all high-resolution images are extracted, calculate the average value of the third similarities, if the average value is greater than the second threshold value, and the license plate number of the suspicious object appears for the first time, determine that the identity authentication of the suspicious object fails.

[0048] This embodiment captures views from four angles, front, back, left, and right, so there will be four corresponding high-resolution images, which are assigned first numbers 1-4. Similarly, if there are 60 third images of different grain trucks stored in the server, they are assigned second numbers 1-60. When calculating, first compare the high-resolution image 1 with the third image 1 to obtain the second similarity between the two. The method for calculating the second similarity is the same as the method for calculating the first similarity, that is, the same convolutional neural network algorithm is used. After comparing all the high-resolution image 1 with the third images 1-60, 60 second similarities are obtained, and the second similarity with the largest value is extracted as the third similarity of the high-resolution image 1; then repeat the process to obtain the third similarity of the high-resolution images 2-4, and calculate the average value of the third similarities of the high-resolution images 1-4. The larger the average value, the greater the probability that the grain truck has appeared before. Finally, identify the license plate number included in the high-resolution image and compare it with the number registered when entering the warehouse, so as to determine whether the grain truck has entered the grain warehouse before.

[0049] In this embodiment, analyzing the second environment video includes the following steps: A first standard image of the grain storage building is captured in advance, and the second environment video is split into multiple fourth images based on a preset time interval. The first standard image and the fourth image are gray-scaled to obtain a second standard image and a fifth image. The fifth image is segmented into multiple building outlines based on a contour segmentation algorithm, and a sixth image containing only one building outline is captured from the fifth image. The sixth image and the second standard image are processed to the same size and subtracted to obtain a difference image. If the difference image has multiple consecutive pixel points greater than a third threshold, it is determined that the grain storage building is abnormal.

[0050] Generally, granary buildings all adopt the same architectural shape, so in actual operation, only the first standard image of a granary building can be retained. This embodiment randomly extracts a fourth image as the analysis object. After extracting the fourth image, the fourth image and the first standard image are first converted into grayscale images to reduce the subsequent calculation amount. Since the fifth image is an aerial image taken by a drone, there may be multiple graphics of the top surfaces of granary buildings. Therefore, the fifth image is processed by a contour segmentation algorithm to obtain the contours of multiple building top surfaces, and then the fifth image is cropped to obtain a sixth image containing only one building contour.

[0051] Finally, each sixth image is processed to the same size as the second standard image, for example, 1280*720, and then the pixel values ​​of the pixels at the same position in the two images are subtracted to obtain a difference image. If the outer surface of the grain storage building is completed, the pixel value of each pixel in the difference image obtained after subtraction should be close to 0. If a region of pixels appears in the difference image with larger pixels, it indicates that the corresponding region may be damaged or have foreign matter, and relevant personnel can be reminded to go and check.

[0052] In this embodiment, segmenting the outline building includes the following steps: Determine edge pixels in the fifth image, and sequentially connect adjacent edge pixels to complete segmentation of the fifth image. Specifically, when determining edge pixels, if there is a target pixel, and adjacent pixels around the target pixel meet a preset condition, the target pixel is defined as an edge pixel, and the preset condition is that a difference in pixel values ​​of adjacent pixels is greater than a fourth threshold.

[0053] Since the fifth image is a grayscale image, each pixel point has only one channel of pixel value and is between 0-255; edge pixel points are pixel points on the boundaries of each contour in the image. When determining, the pixel values ​​of each adjacent pixel point around the pixel point are obtained, that is, the pixel values ​​of pixels in eight directions: up, down, left, right, upper left, lower left, upper right, and lower right. If the pixel value difference between two adjacent pixel points is too large, it is considered that the target pixel point is on the contour boundary; this is because the target pixel point is on the boundary contour, which leads to a large difference in the pixel values ​​of the pixels on both sides. For example, the left side of the target pixel point is the top of the building, and the right side is the ground.

[0054] like Figure 2 As shown, the present invention also provides an agricultural product storage system based on drone safety management, which is used to implement the above-mentioned agricultural product storage method based on drone safety management, and the system includes: A setting module, used to set a target area, divide the target area into multiple sub-areas, record the coverage of each sub-area, assign a unique area ID to each sub-area, set a first cruise path for each sub-area, and set a second cruise path for the target area; The patrol module divides the patrol mission into the first mode and the second mode. After the UAV is started, it obtains the mode of this patrol mission. If the patrol mission is the first mode, the UAV goes to the corresponding sub-areas in sequence according to the area ID included in this patrol mission, and obtains the corresponding first cruise path according to the area ID, and shoots the first environment video within the coverage of the sub-area based on the first cruise path. If the patrol mission is the second mode, the UAV takes a bird's-eye view of the target area based on the second cruise path to obtain the second environment video; The analysis module is used to analyze the first environment video and the second environment video to obtain a first analysis result and a second analysis result respectively. The first analysis result includes the situation of the grain transport vehicle in the grain warehouse, and the second analysis result includes the integrity of the grain warehouse building.

[0055] The above embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

[0056] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for storing agricultural products based on drone safety management, characterized in that: include Set a target area, divide the target area into multiple sub-areas, record the coverage of each sub-area, and assign a unique area ID to each sub-area; A first cruising path is set for each of the sub-areas, and a second cruising path is set for the target area; The patrol mission is divided into the first mode and the second mode, and the mode of the patrol mission is obtained after the drone is started; If the patrol mission is the first mode, the drone goes to the corresponding sub-areas in sequence according to the area ID included in this patrol mission, obtains the corresponding first cruise path according to the area ID, and shoots the first environment video within the coverage of the sub-area based on the first cruise path; If the patrol mission is the second mode, the drone takes a bird's-eye view of the target area based on the second cruise path to obtain a second environment video; The first environmental video and the second environmental video are analyzed to obtain a first analysis result and a second analysis result respectively, wherein the first analysis result includes the situation of the grain transport vehicle in the grain warehouse, and the second analysis result includes the integrity of the grain warehouse building.

2. The method according to claim 1, characterized in that: Obtaining the first analysis result includes the following steps: Splitting the first environment video into a plurality of first images based on a preset time interval, identifying the first images, and acquiring the grain transport vehicle included therein; A second image containing only a single grain truck is cropped from the first image, and the grain truck in each of the second images is analyzed to determine a suspicious object therein, and an area where the suspicious object is located is defined as a suspicious area. The drone goes to the suspicious area, surrounds the suspicious object, and takes high-resolution images from multiple angles; The high-resolution image is analyzed to authenticate the identity of the suspicious object. If the identity authentication fails, a warning message is generated and sent to a preset terminal.

3. The method according to claim 2, characterized in that: Analyzing the grain transport vehicle comprises the following steps: When the grain transport vehicle enters the warehouse, a third image is taken from multiple angles, the third image is stored and a monitoring time is set; After acquiring the second image, the drone acquires the third image which is still within the monitoring time, extracts a bird's-eye view from the third image and defines it as a historical image, calculates a first similarity between the second image and each of the historical images, and marks the grain truck as the suspicious object if the first similarity is greater than a first threshold.

4. The method according to claim 3, characterized in that: Analyzing the high-resolution image comprises the following steps: Assigning a first number to each of the high-resolution images, assigning a second number to each of the third images, extracting the first high-resolution image with the first number, and calculating the second similarities with each of the third images with the second numbers, and extracting the second similarity with the largest value as the third similarity corresponding to the first high-resolution image with the first number; Extract the second high-resolution image of the first number, calculate the second similarity with each of the third images of the second number, and determine the third similarity therefrom, repeat this step until all the high-resolution images are extracted, calculate the average value of the third similarities, and if the average value is greater than a second threshold value and the license plate number of the suspicious object appears for the first time, determine that the identity authentication of the suspicious object has failed.

5. The method according to claim 2, characterized in that: Analyzing the second environment video includes the following steps: A first standard image of the grain storage building is captured in advance, the second environment video is split into multiple fourth images based on the preset time interval, the first standard image and the fourth image are gray-scaled to obtain a second standard image and a fifth image, the fifth image is segmented into multiple building outlines based on a contour segmentation algorithm, a sixth image containing only one building outline is captured from the fifth image, the sixth image and the second standard image are processed to the same size and subtracted to obtain a difference image, and if the difference image has multiple consecutive pixel points whose pixel points are greater than a third threshold, it is determined that the grain storage building is abnormal.

6. The method according to claim 5, characterized in that: Segmenting the outline building comprises the following steps: The edge pixel points in the fifth image are determined, and the edge pixel points that are adjacent in position are sequentially connected to complete the segmentation of the fifth image.

7. The method according to claim 6, characterized in that: Determining the edge pixel point comprises the following steps: If there is a target pixel point, and the neighboring pixel points around the target pixel point meet a preset condition, the target pixel point is defined as the edge pixel point, and the preset condition is that the difference in pixel values ​​of the neighboring pixel points is greater than a fourth threshold.

8. The method according to claim 2, characterized in that: Recognizing the first image comprises the following steps: A picture set containing the grain truck is obtained, and a training data set is obtained after annotation. A YOLO network model is established, the constructed YOLO network model is trained, and the checkpoint file generated during the training process is tested. Finally, the checkpoint file with the highest mAP is selected as the final network model, and the first image is recognized based on the final network model.

9. The method according to claim 8, characterized in that: The final network model includes a backbone network, a Neck, and a prediction output layer. The backbone network adopts the MobileNetv3 network structure. The Neck includes three side output layers and an output layer. The side output layers are the SPP layer, the feature pyramid layer, and the upsampling layer.

10. An agricultural product storage system based on drone safety management, used to implement the method according to any one of claims 1 to 9, characterized in that: A setting module, used to set a target area, divide the target area into multiple sub-areas, record the coverage of each sub-area, assign a unique area ID to each sub-area, set a first cruising path for each sub-area, and set a second cruising path for the target area; The patrol module divides the patrol mission into a first mode and a second mode. After the UAV is started, the mode of the patrol mission is obtained. If the patrol mission is the first mode, the UAV goes to the corresponding sub-areas in sequence according to the area ID included in the patrol mission, and obtains the corresponding first cruise path according to the area ID, and shoots the first environment video within the coverage of the sub-area based on the first cruise path. If the patrol mission is the second mode, the UAV takes a bird's-eye view of the target area based on the second cruise path to obtain the second environment video; The analysis module is used to analyze the first environmental video and the second environmental video to obtain a first analysis result and a second analysis result respectively. The first analysis result includes the situation of the grain transport vehicle in the grain warehouse, and the second analysis result includes the integrity of the grain warehouse building.

Images