A granary inspection robot path planning method and system

Abstract

The present application relates to the technical field of path planning, in particular to a kind of granary inspection robot path planning method and system;The method of the present application first constructs the three-dimensional coordinate system of granary, then robot travels according to preset path and real-time collection, transmission, identifies image, can mark abnormal position and image in time, reduce abnormal missed inspection or delay, improve the timeliness and accuracy of abnormal preliminary judgment;Then, based on multi-frame data identification insect density, predict three-dimensional activity area, then combine insect density, carry out spiral trajectory path planning in insect activity three-dimensional area, determine trajectory offset parameter, improve the accuracy of abnormal area inspection;Finally, with abnormal position as starting point, according to trajectory parameter offset inspection, focus on pest area to strengthen coverage, reduce invalid path, further improve the intelligent accuracy of granary pest area monitoring.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, specifically to a path planning method and system for a grain warehouse inspection robot. Background Technology

[0002] Food security is one of the core foundations for ensuring people's livelihood in the country. As a key facility for large-scale grain storage, grain warehouses are one of the main sources of risk for grain spoilage, quality deterioration, and storage losses due to the breeding and activity of pests inside them. Therefore, regular and precise grain warehouse inspections are a necessary link to ensure the safety of stored grain.

[0003] Traditional grain warehouse inspections rely on manual labor. Limited by the high dust concentration and harsh environment of the enclosed space, manual inspections are not only inefficient and labor-intensive, but also pose health and safety risks. Furthermore, manual inspections depend on experience-based judgment, making them prone to missed inspections and misjudgments, and they cannot generate unified spatial inspection data, hindering the accurate tracking of pest activity areas. Against this backdrop, grain warehouse inspection robots are gradually being applied to grain storage management; however, existing technologies still have many shortcomings that require optimization and cannot meet the demands for high-precision, intelligent inspections.

[0004] Currently available grain warehouse inspection robots mostly adopt a pre-set fixed path planning method, which can only achieve traversal inspection of regular areas and lacks a dynamic response mechanism for abnormal areas. When an abnormality of pests is detected, the existing technology blindly expands the inspection range, resulting in an increase in invalid paths and a decrease in inspection efficiency, or only makes small adjustments near the original path, resulting in insufficient coverage of pest areas. Ultimately, it is difficult to balance the accuracy and efficiency of the inspection and cannot effectively support the refined prevention and control of grain warehouse pest risks. Summary of the Invention

[0005] The purpose of this invention is to provide a path planning method and system for a grain warehouse inspection robot.

[0006] The technical solution of this invention is as follows:

[0007] A path planning method for a grain warehouse inspection robot includes the following steps:

[0008] S1: Construct a three-dimensional coordinate system using a point in the granary as the origin.

[0009] S2: The grain warehouse inspection robot enters the grain warehouse at the predetermined location and travels along a pre-set path. During the journey, it collects images at a set frequency and transmits the collected images to the operation and maintenance platform in real time for real-time image recognition and processing.

[0010] If the image recognition result is no abnormality, continue to travel along the pre-set path;

[0011] If the image recognition result indicates an anomaly, record the current position as the anomaly position and the current image as the anomaly image, and execute S3;

[0012] S3: Based on multiple consecutive time-lapse images of anomalous images and locations, insect density is identified and the three-dimensional region of insect activity is predicted;

[0013] Based on insect density and the last frame image, the trajectory offset direction of the grain warehouse inspection robot in the three-dimensional area of ​​insect activity is predicted; based on the trajectory offset direction and the density level corresponding to insect density, a spiral trajectory path is planned in the three-dimensional area of ​​insect activity to obtain the pitch and spiral angle, which, together with the trajectory offset direction, form the trajectory offset parameters.

[0014] Using abnormal locations as the starting point of new paths, the grain warehouse inspection robot performs path offset within the three-dimensional area where insects are active, based on trajectory offset parameters.

[0015] The insect density identification process is as follows: Multiple consecutive frames of images at abnormal locations are acquired using a camera on a grain warehouse inspection robot. Insect features are extracted from both the abnormal images and the Transformer attention module, and the insect feature sharpness entropy of each frame is calculated. Based on the feature sharpness entropy, the attention weights of each frame are obtained, and all frames are weighted and fused based on these attention weights to obtain a fused insect feature map. After image segmentation, the fused insect feature map undergoes connected component counting and density conversion to obtain the insect density.

[0016] Insect feature sharpness entropy is obtained based on the edge gradient features, texture energy features, and illumination uniformity features of the image.

[0017] The method for obtaining the three-dimensional region of insect activity is as follows: Based on multiple consecutive time-lapse images of abnormal images and abnormal locations, the horizontal offset vector and vertical distribution gradient of the insect's center of gravity are obtained, and the three-dimensional activity trend direction is synthesized; based on the insect density, the dynamic diffusion coefficient is matched; based on the abnormal location, insect density, three-dimensional activity trend direction, dynamic diffusion coefficient, and prediction time, the core parameters of the insect's three-dimensional spatial range represented by an ellipsoid are predicted, including the ellipsoid's center, semi-major axis, semi-minor axis, and height.

[0018] The method for obtaining the trajectory offset direction is as follows: the last frame image is processed by Sobel operator texture extraction to obtain a two-dimensional texture gradient field; based on the two-dimensional texture gradient field and insect density, the equipotential surface corresponding to the critical value of insect density is obtained, and the tangent direction of the equipotential surface is used as the initial trajectory offset direction; based on the robot camera calibration parameters, the initial trajectory offset direction is converted into the initial three-dimensional offset direction in the world coordinate system; it is checked whether the extension direction of the initial three-dimensional offset direction exceeds the three-dimensional area of ​​insect activity; if it does not exceed, the initial three-dimensional offset direction is used as the trajectory offset direction; if it exceeds, the initial three-dimensional offset direction is corrected by boundary projection to obtain the corrected three-dimensional offset direction, which is used as the trajectory offset direction.

[0019] In S2, if the image recognition result indicates an anomaly, the grain warehouse inspection robot obtains the grain warehouse temperature at the abnormal location; the insect density obtained based on the image is used as the initial density; the initial density is calibrated according to the difference between the grain warehouse temperature and the grain warehouse reference temperature to obtain the calibrated insect density, which is used to perform the operation of obtaining the trajectory offset direction and trajectory offset angle.

[0020] Based on the temperature of the grain silo at abnormal locations, the three-dimensional region of predicted insect activity is corrected. Specifically, when obtaining the vertical distribution gradient of the insect's center of gravity, the vertical distribution gradient is obtained by the ratio of the difference between the grain silo temperature and the grain silo reference temperature to the magnitude of the difference between the grain silo temperature and the grain silo reference temperature.

[0021] A path planning system for a grain warehouse inspection robot, used to implement the above-mentioned path planning method for a grain warehouse inspection robot, including:

[0022] The 3D coordinate system generation module is used to construct a 3D coordinate system with a specific point in the grain warehouse as the origin.

[0023] The anomaly detection module is used by the grain warehouse inspection robot to enter the grain warehouse at a predetermined location and travel along a pre-set path. During the travel, it collects images at a set frequency and transmits the collected images to the operation and maintenance platform in real time for real-time image recognition processing. If the image recognition result is no anomaly, it continues to travel along the pre-set path; if the image recognition result is that an anomaly exists, the current position is recorded as the anomaly position and the current image is recorded as the anomaly image, and the robot path offset module is executed.

[0024] The robot path offset module is used to identify insect density and predict the three-dimensional area of ​​insect activity based on multiple consecutive frames of images from abnormal images and abnormal locations. Based on the insect density and the last frame image, it predicts the trajectory offset direction of the grain warehouse inspection robot in the three-dimensional area of ​​insect activity. Based on the trajectory offset direction and the corresponding density level of insect density, it performs spiral trajectory path planning in the three-dimensional area of ​​insect activity to obtain the pitch and spiral angle, which, together with the trajectory offset direction, form the trajectory offset parameters. Taking the abnormal location as the starting point of the new path, the grain warehouse inspection robot performs path offset in the three-dimensional area of ​​insect activity according to the trajectory offset parameters.

[0025] A path planning device for a grain warehouse inspection robot includes a processor and a memory, wherein the processor executes a computer program stored in the memory to implement the above-mentioned path planning method for the grain warehouse inspection robot.

[0026] A computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the above-described path planning method for a grain warehouse inspection robot.

[0027] The beneficial effects of this invention are as follows:

[0028] This invention provides a path planning method for a grain warehouse inspection robot. First, a three-dimensional coordinate system for the grain warehouse is constructed to provide a unified spatial reference for subsequent inspection location and area modeling, fundamentally avoiding spatial data deviation and laying the foundation for accurate inspection positioning. Then, the robot travels along a preset path and collects, transmits, and identifies images in real time, enabling timely marking of abnormal locations and images, reducing missed or delayed anomalies, and improving the timeliness and accuracy of initial anomaly judgment. Next, based on multi-frame data, insect density is identified and three-dimensional activity areas are predicted. Combined with insect density, a spiral trajectory path is planned, and trajectory offset parameters are determined to ensure the trajectory adjustment conforms to the actual distribution of insects, improving the accuracy of anomaly area inspection. Finally, inspection is performed from the anomaly location as the starting point, offsetting according to the trajectory parameters, focusing on pest-prone areas to strengthen coverage, reducing invalid paths, and further improving the intelligent accuracy of grain warehouse pest monitoring. Detailed Implementation

[0029] Example 1

[0030] This embodiment describes a scenario where a submersible grain silo inspection robot inspects the inside of a grain silo. During the inspection, the robot acquires images of the grain inside the silo at a fixed frequency and transmits these images to an operation and maintenance platform for image recognition processing. The platform then converts the processed results into control signals, which are transmitted back to the robot to control its operation. The grain silo inspection robot is an existing mechanical device, such as the grain silo inspection robot design disclosed in Chinese Patent ZL202422995409.3.

[0031] S1: Construct a three-dimensional coordinate system with a point in the granary as the origin of the coordinate system.

[0032] S2: The grain warehouse inspection robot enters the grain warehouse at the predetermined location and travels along the pre-set path. During the travel, it collects images at a set frequency and transmits the collected images to the operation and maintenance platform in real time for real-time image recognition processing (which can be achieved through target detection using the YOLOv8 model). If the image recognition result is no abnormality, it continues to travel along the pre-set path; if the image recognition result is abnormal (insects are present in the image), the current position (the current position of the grain warehouse inspection robot) is marked as an abnormal position, the current image is an abnormal image, and S3 is executed.

[0033] S3: Based on multiple consecutive frames of images from abnormal images and locations, insect density is identified, and the three-dimensional area of ​​insect activity is predicted. Based on the insect density and the last frame image, the trajectory offset direction of the grain warehouse inspection robot in the three-dimensional area of ​​insect activity is predicted. Based on the trajectory offset direction and the corresponding density level of the insects, a spiral trajectory path is planned in the three-dimensional area of ​​insect activity to obtain the pitch and spiral angle, which, together with the trajectory offset direction, form the trajectory offset parameters. The abnormal location is used as the starting point of the new path, and the grain warehouse inspection robot performs path offset in the three-dimensional area of ​​insect activity according to the trajectory offset parameters.

[0034] First, based on multiple consecutive time-lapse images of anomalous images and locations, insect density is identified, and the three-dimensional region of insect activity is predicted.

[0035] The specific steps for identifying insect density are as follows: Using a camera on a grain warehouse inspection robot, multiple consecutive time-lapse images (time intervals of 0.2-0.5 seconds) of abnormal locations are acquired. Insect features (outline, texture, motion blur) are extracted from both the consecutive time-lapse images and the abnormal images using a Transformer attention module. The insect feature sharpness entropy of each frame is calculated (lower entropy values ​​indicate sharper features). Based on the feature sharpness entropy, attention weights are assigned to each frame (lower entropy values ​​indicate higher attention weights). All frames are then weighted and fused based on these attention weights to obtain a denoised fused insect feature map, highlighting clear insect features while suppressing noise. The fused insect feature map is then segmented to obtain a fused segmented image. Finally, the fused segmented image undergoes connected component counting and density conversion processing to obtain the insect density. The calculation formula is as follows: p=N×S 0 / S , r For insect density, N The number of connected components. S 0 The actual area of ​​a single pixel. SThe area of ​​the image acquisition region is defined. Based on a preset threshold, the overall insect density is divided into three levels: low density, medium density, and high density.

[0036] Insect feature sharpness entropy is obtained based on the edge gradient features, texture energy features, and illumination uniformity features of the image, and the calculation formula is as follows:

[0037] ,

[0038] Entropy of insect feature clarity, , , These are the edge gradient distribution entropy, texture energy distribution entropy, and illumination uniformity entropy, respectively. , , These are the edge gradient entropy weights, texture energy entropy weights, and illumination uniformity entropy weights, respectively. This is a correction coefficient for the effective proportion of features. For the effective proportion of features, Offset.

[0039] To reduce the impact of dust in grain warehouse images and improve the accuracy of insect identification, adaptive bilateral filtering and local contrast enhancement processing are applied to the images (abnormal images and multi-frame continuous images of abnormal locations) to obtain enhanced images, which are then used to perform the operations in S3.

[0040] The operational steps for predicting the three-dimensional region of insect activity are detailed below.

[0041] Step 1: Establish a local three-dimensional polar coordinate system with the abnormal location as the origin. Based on the abnormal image and multiple consecutive time-lapse images of the abnormal location, obtain the horizontal offset vector (unit vector, representing the horizontal activity trend) and vertical distribution gradient of the insect's center of gravity, and synthesize the three-dimensional activity trend direction by weighting.

[0042] The method for obtaining the horizontal offset vector is as follows: multiple consecutive frames of images at abnormal locations are processed by the YOLOv8 model to obtain the insect target detection boxes for each frame. The overall centroid coordinates of the insects in each frame are obtained by using a confidence-weighted average. The overall centroid coordinates of all insects are summarized with the frame timestamp as the horizontal axis and the overall centroid coordinates of the insects as the vertical axis. Linear fitting is performed in the x and y directions to obtain the fitting slopes in the x and y directions. The obtained fitting slopes in the x and y directions are used to form the initial horizontal offset vector, which is then normalized by L2 to obtain the horizontal offset vector.

[0043] The vertical gradient distribution can be obtained based on the experience of matching insect species in abnormal graphics with the vertical gradient distribution.

[0044] Step 2: Match the dynamic diffusion coefficient based on insect density. , As the reference diffusion coefficient, ρ is the density response coefficient, and ρ1 is the insect density.

[0045] Step 3: Based on the anomaly location, insect density, three-dimensional activity trend direction, dynamic diffusion coefficient, and prediction time, predict the core parameters of the insect's three-dimensional spatial range represented by an ellipsoid, including the ellipsoid center, semi-major axis (along the activity trend direction). ), short semi-axis (vertical) (Horizontal direction) and height (vertical direction).

[0046] ,

[0047] ,

[0048] ,

[0049] ,

[0050] , , , These are the center of the ellipsoid, the semi-major axis, the semi-minor axis, and the height, respectively. This is an abnormal location. The direction of the three-dimensional activity trend is represented by ρ1, where ρ1 is the insect density. The dynamic diffusion coefficient is... The prediction time can be set according to requirements. , , These represent the major semi-axis, minor semi-axis, and height, which are the baselines for the minimum activity range of stored insects.

[0051] Then, based on the insect density and the last frame image, the trajectory offset direction of the grain warehouse inspection robot in the three-dimensional area of ​​insect activity is predicted. The specific steps are detailed below.

[0052] Step 1: The last frame of the multi-frame sequential images at the anomaly location is processed by Sobel operator texture extraction to obtain a two-dimensional texture gradient field. The gradient value of each pixel in the two-dimensional texture gradient field corresponds to the density of the insect texture, and the gradient direction corresponds to the extension direction of the insect aggregation, providing a basis for the offset direction decision. Based on the two-dimensional texture gradient field and insect density, equipotential surfaces are extracted to obtain the equipotential surfaces corresponding to the critical insect density values. The tangent direction of the equipotential surface is used as the initial trajectory offset direction to avoid offsetting along the gradient direction, which would lead to an excessively large detection range.

[0053] The formula for calculating the critical value of insect density is as follows:

[0054] ,

[0055] This is the critical value for insect density. The volume of the three-dimensional region where insects are active. This is the standard volume.

[0056] Step 2: Based on the robot camera calibration parameters, adjust the initial trajectory offset direction. Convert to the initial 3D offset direction in the world coordinate system ; Verify whether the extension direction of the initial three-dimensional offset direction exceeds the three-dimensional area of ​​insect activity; if it does not exceed, use the initial three-dimensional offset direction as the trajectory offset direction of the inspection robot; if it exceeds, perform boundary projection correction on the initial three-dimensional offset direction to obtain the corrected three-dimensional offset direction, which is used as the trajectory offset direction of the inspection robot.

[0057] The following formula can be used to verify whether the extension direction of the initial three-dimensional offset direction exceeds the three-dimensional region of insect activity:

[0058] ,

[0059] (x, y, z) represents the initial 3D offset direction to be input. , , ( ) represents the coordinates of the center of the ellipsoid corresponding to the three-dimensional region of insect activity. , , These represent the ellipsoid center, semi-major axis, semi-minor axis, and height of the ellipsoid corresponding to the three-dimensional region of insect activity. At that time, the extension direction of the initial three-dimensional offset direction did not exceed the three-dimensional region of insect activity.

[0060] Boundary projection correction can be achieved using the following formula:

[0061] ,

[0062] To correct the three-dimensional offset direction, The initial three-dimensional offset direction, , , , These are the absolute values ​​of the components of the initial three-dimensional offset direction along the x, y, and z axes, respectively.

[0063] Next, based on the trajectory offset direction and the corresponding density level of insects, a spiral trajectory path is planned within the three-dimensional area of ​​insect activity to obtain the pitch and spiral angle. In this embodiment, the specific method is as follows: the spiral direction (clockwise / counterclockwise) is determined by the trajectory offset direction, and the core parameters of the spiral trajectory (pitch and spiral angle) are adjusted by the insect density level.

[0064] The formula for calculating the pitch is as follows:

[0065] ,

[0066] ,

[0067] , , These represent the pitch, the maximum allowable pitch, and the minimum allowable pitch, respectively. , Assign values ​​to the density level, and assign the maximum value to the density level (the higher the density, the higher the density level). The higher the density level, the smaller the pitch (high-density areas require denser spiral tracks to cover), and the lower the density level, the larger the pitch (low-density areas reduce ineffective tracks and lower energy consumption). This is a spatial coefficient representing the activity area, reflecting the horizontal / vertical dimensional ratio of the ellipsoid, allowing the spiral parameters to simultaneously adapt to the three-dimensional morphology of the insect's activity area. , , These represent the center, semi-major axis, semi-minor axis, and height of the ellipsoid corresponding to the three-dimensional region of insect activity.

[0068] The formula for calculating the helix angle is as follows:

[0069] ,

[0070] The helix angle, The base helix angle corresponding to the density level. This is the direction correction factor. This indicates the direction of trajectory offset. For directional fine-tuning of gain.

[0071] Example 2

[0072] Example 2 introduces temperature control based on Example 1. Specifically, in S2 of Example 1, if there is an anomaly in the image recognition, the temperature sensor mounted on the grain warehouse inspection robot obtains the temperature of the grain warehouse at the abnormal location; the insect density obtained based on the image is used as the initial density; the initial density is calibrated according to the difference between the grain warehouse temperature and the grain warehouse reference temperature to obtain the calibrated insect density, which is used to perform the operation of obtaining trajectory offset parameters.

[0073] The formula for calculating the calibrated insect density is as follows:

[0074] ,

[0075] To calibrate insect density, For the initial density, This represents the difference between the grain warehouse temperature and the grain warehouse reference temperature. This is the biological thermal activity coefficient (related to insect species). This is the temperature threshold coefficient. This represents the attenuation coefficient at extreme temperatures.

[0076] Furthermore, based on the temperature of the grain silo at abnormal locations, the three-dimensional region of predicted insect activity is corrected. When obtaining the vertical distribution gradient of the insect's center of gravity, the vertical distribution gradient is obtained by the ratio of the difference between the grain silo temperature and the grain silo reference temperature to the modulus of the difference between the grain silo temperature and the grain silo reference temperature.

[0077] This embodiment also provides a path planning system for a grain warehouse inspection robot, used to implement the above-mentioned path planning method for a grain warehouse inspection robot, including:

[0078] The 3D coordinate system generation module is used to construct a 3D coordinate system with a specific point in the grain warehouse as the origin.

[0079] The anomaly detection module is used by the grain warehouse inspection robot to enter the grain warehouse at a predetermined location and travel along a pre-set path. During the travel, it collects images at a set frequency and transmits the collected images to the operation and maintenance platform in real time for real-time image recognition processing. If the image recognition result is no anomaly, it continues to travel along the pre-set path; if the image recognition result is that an anomaly exists, the current position is recorded as the anomaly position and the current image is recorded as the anomaly image, and the robot path offset module is executed.

[0080] The robot path offset module is used to identify insect density and predict the three-dimensional area of ​​insect activity based on multiple consecutive frames of images from abnormal images and abnormal locations. Based on the insect density and the last frame image, it predicts the trajectory offset direction of the grain warehouse inspection robot in the three-dimensional area of ​​insect activity. Based on the trajectory offset direction and the corresponding density level of insect density, it performs spiral trajectory path planning in the three-dimensional area of ​​insect activity to obtain the pitch and spiral angle, which, together with the trajectory offset direction, form the trajectory offset parameters. Taking the abnormal location as the starting point of the new path, the grain warehouse inspection robot performs path offset in the three-dimensional area of ​​insect activity according to the trajectory offset parameters.

[0081] This embodiment also provides a path planning device for a grain warehouse inspection robot, including a processor and a memory, wherein the processor executes the computer program stored in the memory to implement the above-mentioned path planning method for the grain warehouse inspection robot.

[0082] This embodiment also provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the above-described path planning method for a grain warehouse inspection robot.

[0083] This embodiment provides a path planning method for a grain warehouse inspection robot. First, a three-dimensional coordinate system of the grain warehouse is constructed to provide a unified spatial reference for subsequent inspection location and area modeling, fundamentally avoiding spatial data deviation and laying the foundation for accurate inspection positioning. Then, the robot travels along a preset path and collects, transmits, and identifies images in real time, enabling timely marking of abnormal locations and images, reducing missed or delayed anomalies, and improving the timeliness and accuracy of initial anomaly judgment. Next, based on multi-frame data, insect density is identified and three-dimensional activity areas are predicted. Combined with insect density, a spiral trajectory path is planned to determine trajectory offset parameters, allowing the trajectory adjustment to conform to the actual distribution of insects, improving the accuracy of abnormal area inspection. Finally, inspection is carried out from the abnormal location with the trajectory parameter offset, focusing on pest areas to strengthen coverage, reducing invalid paths, and further improving the intelligent accuracy of grain warehouse pest monitoring.

[0084] While exemplary embodiments of the invention have been described herein, many other variations or modifications conforming to the principles of the invention can be directly determined or derived from the disclosure of this invention without departing from its spirit and scope. Therefore, the scope of the invention should be understood and recognized to cover all such other variations or modifications.

Claims

1. A path planning method for a grain warehouse inspection robot, characterized in that, Includes the following steps: S1: Construct a three-dimensional coordinate system using a point in the granary as the origin. S2: The grain warehouse inspection robot enters the grain warehouse at the predetermined location and travels along a pre-set path. During the journey, it collects images at a set frequency and transmits the collected images to the operation and maintenance platform in real time for real-time image recognition and processing. If the image recognition result is no abnormality, continue to travel along the pre-set path; If the image recognition result indicates an anomaly, record the current position as the anomaly position and the current image as the anomaly image, and execute S3; S3: Based on multiple consecutive time-lapse images of anomalous images and locations, insect density is identified, and the three-dimensional region of insect activity is predicted. The method for obtaining the three-dimensional region of insect activity is as follows: Based on multiple consecutive time-lapse images of anomalous images and locations, the horizontal offset vector and vertical distribution gradient of the insect's center of gravity are obtained, and the three-dimensional activity trend direction is synthesized; Based on insect density, the dynamic diffusion coefficient is matched; Based on anomalous location, insect density, three-dimensional activity trend direction, dynamic diffusion coefficient, and prediction time, the core parameters of the insect's three-dimensional spatial range represented by an ellipsoid are predicted, including the ellipsoid's center, semi-major axis, semi-minor axis, and height. Based on insect density and the last frame image, the trajectory offset direction of the grain warehouse inspection robot in the three-dimensional area of ​​insect activity is predicted; based on the trajectory offset direction and the density level corresponding to insect density, a spiral trajectory path is planned in the three-dimensional area of ​​insect activity to obtain the pitch and spiral angle, which, together with the trajectory offset direction, form the trajectory offset parameters. Using abnormal locations as the starting point of new paths, the grain warehouse inspection robot performs path offset within the three-dimensional area where insects are active, based on trajectory offset parameters.

2. The path planning method for a grain warehouse inspection robot according to claim 1, characterized in that, The operation for insect density identification is as follows: The camera on the grain warehouse inspection robot acquires multiple consecutive frames of images of abnormal locations. The insect features are extracted from the abnormal images and the insect features in each frame by the Transformer attention module. The insect feature sharpness entropy of each frame is calculated. Feature sharpness entropy is obtained based on the edge gradient features, texture energy features, and illumination uniformity features of the image. Based on the feature clarity entropy, the attention weights of each frame image are obtained, and all frame images are weighted and fused based on the attention weights to obtain a fused insect feature map. After image segmentation, the fused insect feature map is processed by connected component counting and density conversion to obtain the insect density.

3. The path planning method for a grain warehouse inspection robot according to claim 1, characterized in that, The method for obtaining the trajectory offset direction is as follows: The last frame image is processed by Sobel operator texture extraction to obtain a two-dimensional texture gradient field; based on the two-dimensional texture gradient field and insect density, the equipotential surface corresponding to the critical value of insect density is obtained, and the tangent direction of the equipotential surface is used as the initial trajectory offset direction. Based on the robot camera calibration parameters, the initial trajectory offset direction is converted into the initial three-dimensional offset direction in the world coordinate system; it is then verified whether the extension direction of the initial three-dimensional offset direction exceeds the three-dimensional area of ​​insect activity. If it does not exceed the limit, the initial three-dimensional offset direction will be used as the trajectory offset direction; If it exceeds the limit, the initial three-dimensional offset direction is corrected by boundary projection to obtain the corrected three-dimensional offset direction, which is used as the trajectory offset direction.

4. The path planning method for a grain warehouse inspection robot according to claim 1, characterized in that, In S2, if the image recognition result indicates an anomaly, the grain warehouse inspection robot obtains the temperature of the grain warehouse at the location of the anomaly. The insect density obtained from the image is used as the initial density. The initial density is calibrated according to the difference between the grain warehouse temperature and the grain warehouse reference temperature to obtain the calibrated insect density, which is used to perform the operation of obtaining the trajectory offset direction and trajectory offset angle.

5. The path planning method for a grain warehouse inspection robot according to claim 4, characterized in that, Based on the temperature of the grain silo at abnormal locations, the three-dimensional region of predicted insect activity is corrected. Specifically, when obtaining the vertical distribution gradient of the insect's center of gravity, the vertical distribution gradient is obtained by the ratio of the difference between the grain silo temperature and the grain silo reference temperature to the modulus of the difference between the grain silo temperature and the grain silo reference temperature.

6. A path planning system for a grain warehouse inspection robot, used to implement the path planning method for a grain warehouse inspection robot as described in claim 1, characterized in that, include; The 3D coordinate system generation module is used to construct a 3D coordinate system with a specific point in the grain warehouse as the origin. The anomaly detection module is used by the grain warehouse inspection robot to enter the grain warehouse at a predetermined location and travel along a pre-set path. During the travel, it collects images at a set frequency and transmits the collected images to the operation and maintenance platform in real time for real-time image recognition processing. If the image recognition result is no anomaly, it continues to travel along the pre-set path; if the image recognition result is that an anomaly exists, the current position is recorded as the anomaly position and the current image is recorded as the anomaly image, and the robot path offset module is executed. The robot path offset module is used to identify insect density and predict the three-dimensional area of ​​insect activity based on multiple consecutive frames of images from abnormal images and abnormal locations. Based on the insect density and the last frame image, it predicts the trajectory offset direction of the grain warehouse inspection robot in the three-dimensional area of ​​insect activity. Based on the trajectory offset direction and the corresponding density level of insect density, it performs spiral trajectory path planning in the three-dimensional area of ​​insect activity to obtain the pitch and spiral angle, which, together with the trajectory offset direction, form the trajectory offset parameters. Taking the abnormal location as the starting point of the new path, the grain warehouse inspection robot performs path offset in the three-dimensional area of ​​insect activity according to the trajectory offset parameters.

7. A path planning device for a grain warehouse inspection robot, characterized in that, It includes a processor and a memory, wherein the processor executes a computer program stored in the memory to implement the grain warehouse inspection robot path planning method as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the grain warehouse inspection robot path planning method as described in any one of claims 1-5.