Intelligent path planning abnormity early warning system based on animal husbandry inspection robot

Abstract

The invention provides an intelligent path planning abnormity early warning system based on a livestock inspection robot, and relates to the technical field of robot intelligence, and the system comprises a monitoring module which is used for collecting environment data in real time and monitoring a dynamic obstacle in a driving process according to a preset inspection path through a laser radar and a visual sensor which are disposed on the livestock inspection robot; and the recording module is used for immediately triggering a path re-planning mechanism to obtain a new obstacle avoidance path when it is monitored that the robot cannot run according to the original path due to the dynamic obstacle, and recording preset monitoring points missed due to path deviation. According to the invention, high precision and intelligence of animal husbandry inspection are realized.

Description

Technical Field

[0001] This invention relates to the field of robot intelligence technology, and in particular to an anomaly early warning system based on intelligent path planning for livestock inspection robots. Background Technology

[0002] With the development of large-scale animal husbandry, traditional manual inspections suffer from problems such as high labor costs, incomplete coverage, and delayed response to anomalies. Although existing livestock inspection robots can achieve autonomous path travel, they can only perform basic obstacle avoidance when faced with dynamic obstacles such as livestock gathering in the pen. They lack the ability to identify blind spots and provide risk warnings after path deviation, making it difficult to meet the needs of refined farming.

[0003] A dairy farm with a thousand cows had deployed a routine inspection robot for daily inspections. On one occasion, the robot encountered a dynamic obstacle near the feed feeding point where multiple cows were gathered. After avoiding the obstacle, it deviated from its original preset path. Because it did not record the missed monitoring points and lacked spatial topology analysis capabilities, it failed to detect the missed water level monitoring in the water trough in that area. As a result, some cows experienced a decrease in milk production due to insufficient water. This case exposed the lack of spatial topology construction capabilities with key facilities such as water troughs and feed points as nodes. It was unable to identify blind spots in the inspection after obstacle avoidance through wavefront diffusion, and no corresponding risk assessment and closed-loop verification mechanism was established. Consequently, the monitoring omissions caused by path deviations were difficult to detect and remedy in a timely manner. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide an anomaly early warning system based on intelligent path planning of livestock inspection robots, so as to achieve high precision and intelligence in livestock inspection.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: The first aspect is an anomaly early warning system based on intelligent path planning for livestock inspection robots, including: The monitoring module is used to collect environmental data in real time and monitor dynamic obstacles while the livestock inspection robot travels along the preset inspection path using the lidar and vision sensors configured on it. The recording module is used to immediately trigger the path replanning mechanism when a dynamic obstacle is detected that prevents the robot from traveling along the original path, obtain a new obstacle avoidance path, and record the preset monitoring points missed due to path deviation. The calculation module is used to construct a spatial topology based on the recorded path offset information and missed monitoring point data, using the midpoint of the water trough, the midpoint of the feed feeding point, and the midpoint of the ventilation equipment in the livestock pen, and obtain a grid map accordingly. On the grid map, wavefront diffusion is performed starting from the robot's original path and the new obstacle avoidance path to identify grid areas not covered by diffusion, and wavefront propulsion resistance is extracted as path analysis parameters. The assessment module is used to automatically identify blind spots and corresponding monitoring points caused by dynamic obstacles based on path analysis parameters, and to initiate anomaly risk assessment to obtain the anomaly risk assessment results. The acquisition module is used to obtain the location of the potential hazard area, the expected missed monitoring points, and the risk level classification warning information when the abnormal risk assessment results indicate that there is a potential hazard area. The verification module is used to notify the zookeeper to reschedule the robot to the relevant area for verification and inspection based on the graded early warning information, so as to complete the replanning of the route.

[0006] Furthermore, using the lidar and vision sensors equipped on the livestock inspection robot, environmental data is collected in real time and dynamic obstacles are monitored as the robot travels along a preset inspection path, including: It receives LiDAR point cloud data and visual sensor image data uploaded by the robot, and performs spatiotemporal registration and fusion processing on the multi-source sensor data to obtain fused environmental perception data. Based on the fused environmental perception data, a real-time dynamic map of the robot's surrounding environment is obtained; the real-time dynamic map is then compared with a preset static environmental map to identify dynamic objects in the environment. The system continuously tracks identified dynamic objects and calculates their speed and direction. Based on the speed and direction of the dynamic objects, it predicts the future spatial intersection between the dynamic objects and the robot's preset inspection path. When a spatial intersection is predicted and the intersection time is less than a preset threshold, the system confirms that a dynamic obstacle has been detected.

[0007] Furthermore, when a dynamic obstacle is detected that prevents the robot from following its original path, a path replanning mechanism is immediately triggered to obtain a new obstacle avoidance path, and preset monitoring points missed due to path deviation are recorded, including: After confirming the detection of dynamic obstacles, the path planning algorithm is invoked. Starting from the robot's current position and with the goal of resuming the execution of the preset inspection task, a local path replanning is performed in the global grid map to obtain a new obstacle avoidance path. The new obstacle avoidance path is spatially compared with the preset inspection path, and the offset area between the two paths is calculated. Within the offset area, the preset monitoring points covered by the original preset inspection path but not covered by the new obstacle avoidance path are identified, and the preset monitoring points that are not covered are marked as missed monitoring points. The new obstacle avoidance path, path offset information, and the missing monitoring point data are associated to generate a structured path offset record.

[0008] Furthermore, based on the recorded path offset information and missed monitoring point data, a spatial topology is constructed using the midpoints of the water troughs, feed feeding points, and ventilation equipment within the livestock pen, and a raster map is obtained accordingly, including: Based on the geospatial information in the path offset records, the spatial coordinates of the midpoints of all water troughs, feed feeding points and ventilation equipment are extracted from the predefined livestock pen structure layout database. By extracting the midpoint coordinates of each facility as topological nodes, and determining the connection relationship between nodes based on the actual passage layout within the livestock pen, a topological network representing the spatial connectivity between facilities is constructed. The constructed topology network is used as input and projected onto a two-dimensional Cartesian coordinate plane. The plane is then divided into grids using a preset resolution to obtain a two-dimensional raster map with topological constraints. The geographic coordinates of the original preset inspection path and the new obstacle avoidance path in the path offset record are mapped onto a two-dimensional raster map with topological constraints to obtain a path label map containing two path raster sequences.

[0009] Furthermore, on the grid map, wavefront diffusion is performed starting from the robot's original path and the new obstacle avoidance path to identify grid areas not covered by the diffusion, and wavefront propulsion resistance is extracted as path analysis parameters, including: Based on the path-marked map, the grid sequence of the original preset inspection path and the new obstacle avoidance path is initialized as the starting wave source for wavefront diffusion. The wavefront diffusion algorithm is executed starting from the initial wave source. During the diffusion process, the wavefront arrival value of each grid is recorded, and the corresponding propulsion drag coefficient is calculated based on the channel width, turning angle and obstacle distribution of the grid. Based on the relationship between wavefront arrival value and a preset threshold, grid areas whose arrival value exceeds the threshold and are not covered by the wavefront are identified and marked as potential inspection blind spots. Based on the propulsion drag coefficient and wavefront diffusion path, the average drag intensity and maximum drag value in each diffusion direction are extracted to obtain path analysis parameters for evaluating path reachability.

[0010] Furthermore, based on path analysis parameters, blind spots in inspections caused by dynamic obstacles and corresponding monitoring points are automatically identified, and anomaly risk assessment is initiated to obtain the anomaly risk assessment results, including: Spatial correlation matching is performed on the potential inspection blind spots marked in the path analysis parameters and the dataset of missed monitoring points to identify the actual inspection blind spots caused by dynamic obstacles and the associated monitoring points. Based on the confirmed actual inspection blind spots, the baseline risk value of each blind spot is calculated by combining the propulsion resistance coefficient in the path analysis parameters and the weight of the key facility type associated with the blind spot. The baseline risk value is used to make a weighted correction by introducing a set of environmental impact factor weights, resulting in a weighted comprehensive risk value. Based on the weighted and corrected comprehensive risk value, the risk level is determined using the preset risk level classification rules, resulting in an abnormal risk assessment result that includes the risk level, precise blind spot location, and a list of associated monitoring points.

[0011] Furthermore, when the abnormal risk assessment results indicate the existence of a potential hazard area, the system provides graded early warning information, including the location of the hazard area, the expected missed monitoring points, and the risk level, including: Analyze the abnormal risk assessment results and extract key data such as risk level, blind spot location, and list of associated monitoring points; Based on the extracted key data and combined with the predefined early warning level classification standards, the early warning level corresponding to each potential hazard area is determined; By associating and integrating the warning level with the corresponding spatial coordinates of the hidden danger area and the list of expected missed monitoring points, a structured set of warning data is obtained; Based on the structured early warning data set, graded early warning information is obtained, including the location of potential hazard areas, the expected missed monitoring points, and the risk level.

[0012] Furthermore, based on the tiered early warning information, the zookeeper is notified to reschedule the robot to re-examine the relevant areas, thereby completing the replanning of the route, including: The hierarchical early warning information is analyzed, and the spatial coordinate set of the hidden danger area, the risk level data, and the list of expected missed monitoring points are extracted from the hierarchical early warning information. Based on the extracted spatial coordinate set of the hidden danger area and the risk level data, a robot scheduling suggestion scheme for the review area ranking is obtained through a priority evaluation algorithm. The robot scheduling suggestion scheme is integrated with the list of expected missed monitoring points to construct a structured task instruction set that includes the inspection point sequence and review priority; Based on the structured task instruction set and combined with the real-time environmental grid map, the optimal review path sequence for the robot to go to each potential hazard area is planned and generated. The optimal review path sequence and the corresponding structured task instruction set are sent to the target robot to perform the review and inspection task; The system receives on-site verification data returned by the robot after performing the verification and inspection task, verifies the status of the hidden danger area based on the verification data, updates the environmental database, and completes the dynamic optimization of the path planning parameters.

[0013] In a second aspect, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to execute the system.

[0014] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, performs the system.

[0015] The above-described solution of the present invention has at least the following beneficial effects: Because this invention employs multi-source data fusion of lidar and visual sensors to perceive dynamic obstacles, constructs a spatial topology structure based on the midpoints of drinking troughs, feed feeding points, and ventilation equipment within livestock pens to generate a constrained grid map, identifies uncovered grid areas and extracts path analysis parameters through wavefront diffusion algorithms, conducts risk classification assessment by combining facility type weights and environmental factors, and relies on a closed-loop verification mechanism to schedule robots for supplementary inspections and dynamically optimize path planning parameters, it overcomes the technical problems of existing livestock inspection robots in identifying blind spots after obstacle avoidance, being unable to trace missed monitoring points, lacking quantitative risk assessment, and having closed-loop remedial measures. This achieves full-coverage and precise livestock inspection, quantifiable early warning of abnormal risks, and timely verification and supplementary inspection of missed areas, improving the inspection efficiency and refined management level of livestock farms and reducing breeding risks caused by inspection omissions. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an anomaly early warning system based on intelligent path planning of livestock inspection robots provided in an embodiment of the present invention.

[0017] Figure 2 This is a flowchart illustrating the process of an intelligent path planning anomaly early warning system for livestock inspection robots, based on path analysis parameters, automatically identifying blind spots and corresponding monitoring points caused by dynamic obstacles, and initiating anomaly risk assessment to obtain the anomaly risk assessment results. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0019] like Figure 1 As shown, an embodiment of the present invention proposes an anomaly early warning system based on intelligent path planning of livestock inspection robots, including: The monitoring module is used to collect environmental data in real time and monitor dynamic obstacles while the livestock inspection robot travels along the preset inspection path using the lidar and vision sensors configured on it. The recording module is used to immediately trigger the path replanning mechanism when a dynamic obstacle is detected that prevents the robot from traveling along the original path, obtain a new obstacle avoidance path, and record the preset monitoring points missed due to path deviation. The calculation module is used to construct a spatial topology based on the recorded path offset information and missed monitoring point data, using the midpoint of the water trough, the midpoint of the feed feeding point, and the midpoint of the ventilation equipment in the livestock pen, and obtain a grid map accordingly. On the grid map, wavefront diffusion is performed starting from the robot's original path and the new obstacle avoidance path to identify grid areas not covered by diffusion, and wavefront propulsion resistance is extracted as path analysis parameters. The assessment module is used to automatically identify blind spots and corresponding monitoring points caused by dynamic obstacles based on path analysis parameters, and to initiate anomaly risk assessment to obtain the anomaly risk assessment results. The acquisition module is used to obtain the location of the potential hazard area, the expected missed monitoring points, and the risk level classification warning information when the abnormal risk assessment results indicate that there is a potential hazard area. The verification module is used to notify the zookeeper to reschedule the robot to the relevant area for verification and inspection based on the graded early warning information, so as to complete the replanning of the route.

[0020] In this embodiment of the invention, because the invention employs the technical means of collaboratively collecting environmental data and monitoring dynamic obstacles using lidar and visual sensors, triggering path replanning upon encountering obstacles and simultaneously recording missed monitoring points, constructing a spatial topology structure based on the midpoint of key livestock facilities and generating a grid map, identifying uncovered areas through wavefront diffusion analysis, completing blind spot identification and risk assessment based on path analysis parameters, obtaining graded early warning information, and scheduling robots for supplementary inspection and replanning through a closed-loop verification mechanism, it overcomes the technical problems of existing livestock inspection robots easily creating blind spots after obstacle avoidance, difficulty in tracing missed monitoring points, and lack of targeted risk assessment and closed-loop remedial measures. This achieves completeness and accuracy of livestock inspection coverage, early warning and quantitative control of abnormal risks, timely supplementary inspection of missed areas, and dynamic path optimization, significantly improving the intelligence level and refined management efficiency of livestock farm inspections, and reducing breeding safety hazards caused by inspection omissions.

[0021] In a preferred embodiment of the present invention, the livestock inspection robot, equipped with a lidar and vision sensor, collects environmental data in real time and monitors dynamic obstacles while traveling along a preset inspection path, including: The system receives LiDAR point cloud data and visual sensor image data uploaded by the robot, and performs spatiotemporal registration and fusion processing on the multi-source sensor data to obtain fused environmental perception data. Specifically, when the livestock inspection robot travels along a preset inspection path, it simultaneously activates LiDAR and visual sensors to continuously collect information about the surrounding environment. LiDAR captures three-dimensional spatial distance data of facilities such as feed feeding points and water troughs, as well as dairy cows, while visual sensors acquire on-site image information. After receiving these two types of data, the system first calibrates the timestamps and spatial coordinate systems of the multi-source sensors to eliminate time differences and positional deviations in data collection. Then, it removes redundant information and retains effective environmental features through complementary data fusion processing, ultimately forming comprehensive and accurate fused environmental perception data.

[0022] Based on the fused environmental perception data, a real-time dynamic map of the robot's surrounding environment is obtained. The real-time dynamic map is then compared with a preset static environmental map to identify dynamic objects in the environment. Specifically, this involves: constructing a real-time dynamic environmental map of the robot's surroundings based on the fused environmental perception data. This map clearly shows the current distribution of dairy cows, the location of facilities, and the condition of passageways. Simultaneously, a pre-stored static map of the dairy farm is retrieved. This static map includes information such as the fixed layout of livestock pens, the fixed locations of water troughs, feed points, and ventilation equipment. By comparing the real-time dynamic map with the preset static map area by area, objects whose positions have changed are filtered out, and dynamic objects such as dairy cows are accurately identified, avoiding misjudging fixed facilities as obstacles.

[0023] The system continuously tracks identified dynamic objects and calculates their speed and direction. Based on these speeds and directions, it predicts the future spatial intersection between the dynamic objects and the robot's preset inspection path. When spatial intersection is predicted and the intersection time is below a preset threshold, a dynamic obstacle is confirmed to have been detected. Specifically, this includes continuously monitoring and tracking the identified dynamic objects, i.e., the herd of cows, and calculating the herd's speed and direction of travel by analyzing multiple frames of continuously collected data. Combining the coordinate information of the robot's preset inspection path, the system simulates and extrapolates the future positional relationship between the herd of cows and the robot's inspection path to determine whether they will overlap spatially. If the extrapolation results show spatial overlap and the expected overlap time is less than a set threshold, a dynamic obstacle that will affect the inspection path is officially confirmed to have been detected.

[0024] In this embodiment of the invention, because the invention employs the following technical means: receiving lidar point cloud data and visual sensor image data and performing spatiotemporal registration and fusion processing; generating a real-time dynamic map based on the fused data and differentially identifying dynamic objects with a preset static map; continuously tracking dynamic objects and calculating their speed and direction; predicting the future spatial intersection of the object and the robot's preset path and using whether the intersection time is lower than a preset threshold as the dynamic obstacle confirmation standard, the invention effectively overcomes the technical problems of existing livestock inspection robots having low accuracy in identifying dynamic obstacles in complex breeding environments, being prone to misjudgment or omission, and having difficulty predicting path conflicts in advance. Thus, it achieves real-time, accurate, and efficient monitoring of dynamic obstacles in livestock farms.

[0025] In a preferred embodiment of the present invention, when a dynamic obstacle is detected that prevents the robot from traveling along the original path, a path replanning mechanism is immediately triggered to obtain a new obstacle avoidance path, and preset monitoring points missed due to path deviation are recorded, including: After confirming the detection of dynamic obstacles, the path planning algorithm is invoked. Starting from the robot's current position and aiming to resume the preset inspection task, a local path replanning is performed on the global grid map to obtain a new obstacle avoidance path. Specifically, when the livestock inspection robot is traveling near the feed dispensing point, after confirming the detection of a dynamic obstacle formed by multiple dairy cows gathered through the fusion data of LiDAR and visual sensors, the preset path replanning mechanism is immediately triggered. The robot prioritizes the use of a path planning algorithm optimized for the livestock farm scenario. This algorithm has been pre-adapted to the environmental characteristics of dairy farms, such as narrow passages, dense facilities, and unpredictable livestock activities, ensuring both obstacle avoidance and inspection efficiency. The exact location three meters east of the current feed dispensing point is set as the path starting point, accurately pinpointing the corresponding coordinates on the global grid map. The original planned path is then replanned. The continuous inspection tasks, such as monitoring the water quality of the No. 3 water trough and checking the operation status of the No. 5 ventilation equipment, were set as the final target points to ensure that the replanned path did not deviate from the overall inspection process. Relying on a global grid map containing information such as the complete layout of the dairy farm pens, the precise coordinates of all water troughs, feed points, and ventilation equipment, as well as the width and turning angle of each passage, the map, with a grid resolution of 0.5 meters, can clearly show every fixed facility and passable area. During the calculation process, the algorithm strictly followed the core constraint of not touching the dairy cow gathering area, while also considering the robot's own turning radius and driving speed limitations to avoid planning dangerous paths in overly steep turns or narrow passages. After multiple rounds of path scheme comparison and optimization, a new obstacle avoidance path was finally planned, which detours westward from the current position to avoid the dairy cow gathering area and then turns southward, accurately connecting to the subsequent inspection task nodes.

[0026] The new obstacle avoidance path is spatially compared with the preset inspection path to calculate the offset area between the two paths. Within the offset area, preset monitoring points covered by the original preset inspection path but not covered by the new obstacle avoidance path are identified and marked as missed monitoring points. Specifically, this involves a segment-by-segment, grid-by-grid omnidirectional spatial position comparison between the newly planned obstacle avoidance path and the robot's initial preset inspection path. During the comparison, the coordinate overlap of the two paths is verified point by point using the coordinate system of the global grid map as a reference. By calculating the deviation distance and angle of continuous path segments, the completely overlapping areas, partially overlapping areas, and completely non-overlapping offset areas of the two paths are accurately delineated, and the starting coordinates, ending coordinates, and coverage range of the offset areas are determined. Within the identified offset areas, a list of all preset monitoring points planned within the area of ​​the preset inspection path is retrieved. The monitoring includes various types such as water trough level monitoring, feed residue monitoring, and ground temperature and humidity monitoring. Each monitoring point has a clear location coordinate and effective monitoring coverage radius. The effective coverage area of ​​each monitoring point is checked one by one to see if it intersects with the new obstacle avoidance path. Monitoring points whose effective monitoring range was completely covered by the original preset inspection path are identified, but which are completely outside the monitoring range of the new obstacle avoidance path due to the obstacle avoidance. This includes the water level monitoring point of the No. 2 water trough in the area, which was not monitored in the background case. The effective coverage radius of this monitoring point is 1.5 meters. The original path passed right through its coverage area, but the shortest distance between the new obstacle avoidance path and this monitoring point is 3 meters, completely exceeding the monitoring range. The uncovered monitoring points are uniformly marked as missed monitoring points, and the type, number, specific location coordinates, and corresponding monitoring items of each missed monitoring point are recorded simultaneously to ensure that no monitoring points are missed.

[0027] The new obstacle avoidance path, path offset information, and the data of the missed monitoring points are associated to generate a structured path offset record. Specifically, this includes: deeply associating the complete coordinate sequence, path segmentation information, and driving speed planning data of the newly generated obstacle avoidance path with the previously calculated start and end coordinates, coverage area, and deviation direction of the path offset area, as well as detailed data such as the type, number, location coordinates, and monitoring items of each marked missed monitoring point; during the association process, the timestamp of the path change is used as a unified benchmark to accurately match the offset area with the corresponding road segment of the new obstacle avoidance path, and then the location coordinates of each missed monitoring point are mapped to the specific segment of the offset area, ensuring that the three types of information are completely corresponding in spatial location and time node, forming a complete data chain; according to the unified data format preset by the dairy farm management system, the associated information is integrated and sorted to obtain a structured path offset record containing the specific reason for the path change, the location range and number of obstacles, the detailed coordinate range and coverage area of ​​the offset area, the complete information of each missed monitoring point, the start time and completion time of the path change, and the complete coordinate sequence of the new obstacle avoidance path.

[0028] In this embodiment of the invention, because the invention employs the following technical means: after confirming dynamic obstacles, it calls a path planning algorithm; starting from the robot's current position and aiming to restore the preset inspection task, it performs local path replanning on the global grid map to obtain a new obstacle avoidance path; it spatially compares the new obstacle avoidance path with the preset inspection path to calculate the offset area and identifies and marks the missed monitoring points covered by the original path but not covered by the new path; and it associates the new obstacle avoidance path, path offset information, and missed monitoring point data to generate a structured path offset record. Therefore, this invention overcomes the technical problems of existing livestock inspection robots, such as difficulty in accurately defining the path offset range after obstacle avoidance replanning, inability to identify missed preset monitoring points, and lack of systematic recording of path offset-related information, which leads to difficulties in subsequent traceability. Thus, it achieves accurate planning of obstacle avoidance paths, clear definition of offset areas, and accurate marking and structured retention of missed monitoring points, ensuring the continuity of livestock inspection tasks and the integrity of monitoring coverage.

[0029] In a preferred embodiment of the present invention, based on recorded path offset information and missed monitoring point data, a spatial topology is constructed using the midpoints of the water troughs, feed feeding points, and ventilation equipment within the livestock pen, and a raster map is obtained accordingly, including: Based on the geospatial information in the path offset records, the midpoints of all water troughs, feed feeding points, and ventilation equipment are extracted from a predefined livestock pen structure layout database. Specifically, this involves: retrieving previously generated structured path offset records and extracting geospatial information including the pen partition number of the offset area and the location range of any missed monitoring points; locating the corresponding No. 3 livestock pen; and then filtering and extracting the midpoint spatial coordinates of all key facilities within No. 3 from the predefined livestock pen structure layout database. The database stores detailed information on facilities in all livestock pens, including the unique number, installation location, and dimensions of each water trough, feed feeding point, and ventilation equipment. The midpoint coordinates of 12 water troughs, 8 feed feeding points, and 6 ventilation equipment within each pen are accurately extracted, with each coordinate accurate to the centimeter level. By using the extracted coordinates of the midpoints of each facility as topological nodes, and determining the connection relationships between nodes based on the actual passage layout within the livestock pen, a topological network characterizing the spatial connectivity between facilities is constructed. Specifically, this involves: using all 26 midpoint coordinates extracted in the first step as core nodes of the topological network; then retrieving the actual passage layout data for livestock pen No. 3; the data includes detailed information such as the direction, width, turning angle, and presence of intersections of the passages; the system uses this information to determine the actual connectivity between each node; for example, there is a 2.5-meter-wide straight passage without any obstructions between the midpoint of water trough No. 1 and the midpoint of feed feeding point No. 2, indicating a direct connection between them; however, there is no direct passage between the midpoint of water trough No. 5 and the midpoint of ventilation equipment No. 3, separated by a fence, indicating no direct connection between them; and confirming the connection status between all nodes one by one according to the actual passage layout, ultimately constructing a topological network that accurately characterizes the spatial connectivity between key facilities within livestock pen No. 3.

[0030] The constructed topology network is used as input and projected onto a two-dimensional Cartesian coordinate plane. The plane is then divided into grids at a preset resolution to obtain a two-dimensional raster map with topological constraints. Specifically, this involves: using the constructed topology network of livestock pen No. 3 as the core input data and projecting it onto a pre-defined two-dimensional Cartesian coordinate plane; the coordinate system uses the southwest corner vertex of livestock pen No. 3 as the origin, the X-axis as the direction parallel to the long side of the pen, and the Y-axis as the direction parallel to the short side, ensuring that the spatial position of the topology network completely corresponds to the actual pen layout; then, the two-dimensional plane is uniformly divided into grids at a preset grid resolution of 0.5 meters, with each grid representing a 0.5-meter by 0.5-meter area in the real space; during the division process, combined with the connectivity constraints of the topology network, the grids corresponding to passages are marked as passable grids, while the grids corresponding to the facility body fences, etc., are marked as impassable grids, ultimately forming a two-dimensional raster map of livestock pen No. 3 with clear topological constraints.

[0031] The geographic coordinates of the original preset inspection path and the new obstacle avoidance path in the path offset record are mapped onto a topologically constrained two-dimensional raster map to obtain a path-annotated map containing two path raster sequences. Specifically, this involves: extracting the complete geographic coordinate sequences of the original preset inspection path and the new obstacle avoidance path from the path offset record. The original path coordinate sequence contains continuous coordinate points from the starting point to the target point, while the new obstacle avoidance path coordinate sequence contains the coordinates of all turning nodes during the detour. These two sets of coordinate sequences are mapped one by one onto a topologically constrained two-dimensional raster map of the No. 3 livestock pen according to the two-dimensional Cartesian coordinate system parameters set in step three, with each coordinate point precisely corresponding to a specific raster in the map. Then, all the rasteres covered by each path are arranged sequentially to form two complete path raster sequences, with the original path raster sequence covering 120 rasteres and the new obstacle avoidance path raster sequence covering 156 rasteres. These two path raster sequences are marked on the raster map with different colors to obtain a path-annotated map that includes the path coverage area and spatial differences.

[0032] In this embodiment of the invention, because the invention uses geospatial information based on path offset records to extract the spatial coordinates of all midpoints of water troughs, feed feeding points, and ventilation equipment from a predefined livestock pen structure layout database, and uses these facility midpoint coordinates as topological nodes, and determines the connection relationship between nodes according to the actual livestock pen passage layout to construct a topological network representing the spatial connectivity of facilities, the topological network is projected onto a two-dimensional Cartesian coordinate system plane and divided into a grid with a preset resolution to obtain a two-dimensional raster map with topological constraints, and the geographic coordinates of the original preset inspection path and the new obstacle avoidance path are mapped onto the raster map to generate a path label map containing two path raster sequences, the invention overcomes the technical problems of existing technologies that lack the ability to construct spatial topology with key facilities of the livestock farm as nodes, lack of scenario-based topological constraints in the generated raster map, and inability to clearly present the spatial relationship between the two paths before and after obstacle avoidance, resulting in the difficulty in accurately identifying inspection blind spots, and thus achieves accurate matching between the raster map and the actual facility layout and passage distribution of the livestock farm, making the spatial positional relationship between the original path and the new obstacle avoidance path intuitively visible.

[0033] In a preferred embodiment of the present invention, wavefront diffusion is performed on the grid map, starting from the robot's original path and the new obstacle avoidance path, to identify grid areas not covered by the diffusion, and wavefront propulsion resistance is extracted as a path analysis parameter, including: Based on the path-marked map, the grid sequences of the original preset inspection path and the new obstacle avoidance path are initialized as the starting source of wavefront diffusion. Specifically, this includes: retrieving the path-marked map of livestock pen No. 3, which clearly marks the grid sequences covered by the original preset inspection path and the new obstacle avoidance path; the grid sequence of the original path contains 120 consecutive grids from the feed feeding point through the No. 2 water trough to the ventilation equipment, and the grid sequence of the new obstacle avoidance path contains 156 consecutive grids around the west passage; the grid sequences of these two paths are marked to clarify the specific location of each grid on the map, and the marked grid sequences are uniformly set as the starting source of wavefront diffusion.

[0034] The wavefront diffusion algorithm is executed starting from the initial wave source. During the diffusion process, the wavefront arrival cost of each grid cell is recorded. The corresponding propulsion drag coefficient is calculated based on the channel width, turning angle, and obstacle distribution of the grid cell. Specifically, the wavefront diffusion algorithm is executed from the calibrated initial wave source, gradually spreading outwards to adjacent passable grid cells. During the diffusion process, the cost of wavefront arrival for each grid cell is recorded in real time, increasing with the diffusion distance. For example, the cost is 1 for a grid cell from the initial wave source, 2 for a grid cell, and so on. Simultaneously, the propulsion drag coefficient is calculated based on the actual environment of each grid cell: if the grid cell is located in a 2.5-meter-wide main channel, the drag coefficient is set to 1.0; if it is located in a 1.2-meter-wide narrow channel, the drag coefficient increases to 1.8; if it is located at a 90-degree sharp turn, the drag coefficient rises to 2.5; if there are temporary obstacles nearby, such as a small amount of scattered feed, the drag coefficient increases by an additional 0.5. By considering all these factors, a unique propulsion drag coefficient is assigned to each grid cell.

[0035] Based on the relationship between wavefront arrival cost and a preset threshold, grid areas with a cost exceeding the threshold and not covered by the wavefront are identified and marked as potential inspection blind spots. Specifically, based on a preset threshold for wavefront arrival cost and considering the inspection requirements of pen number three, the threshold is set to 5, meaning areas where the wavefront spread exceeds 5 grid squares are considered outside the effective inspection coverage. The wavefront arrival cost of all grids is compared with this threshold, and grid areas with a cost exceeding 5 and not covered by wavefront spread are selected. In the scenario described in the background case, the cost of three grids around water trough number two reaches 6 and is not covered by the wavefront of either the original path or the new obstacle avoidance path; this area is marked as a potential inspection blind spot.

[0036] Based on the propulsion drag coefficient and wavefront diffusion path, the average drag intensity and maximum drag value in each diffusion direction are extracted to obtain path analysis parameters for evaluating path accessibility. Specifically, this includes: based on the propulsion drag coefficient of each grid and the path formed by wavefront diffusion, data are statistically analyzed according to the four main diffusion directions: east, south, west, and north. In the east direction, the diffusion path passes through 3 narrow channels and 1 sharp turn, with an average drag intensity of 2.1 and a maximum drag value of 2.5 at the sharp turn. In the west direction, the diffusion path is mainly through the main channel, with an average drag intensity of 1.2 and a maximum drag value of 1.5. In the south and north directions, due to their proximity to the fence, the diffusion range is smaller, with average drag intensities of 1.8 and 1.6 respectively, and a maximum drag value of 2.0 for both. The average drag intensity and maximum drag value in each direction are summarized to obtain path analysis parameters for evaluating regional path accessibility.

[0037] In this embodiment of the invention, because the invention adopts the technical means of initializing the grid sequence of the original preset inspection path and the new obstacle avoidance path as the starting wave source based on the path annotation map, executing the wave front diffusion algorithm from the starting wave source and recording the wave front arrival cost of each grid and calculating the propulsion resistance coefficient based on factors such as channel width, identifying potential inspection blind spots that are not covered based on the cost value and preset threshold, and extracting the average resistance intensity and maximum resistance value in each direction based on the propulsion resistance coefficient and diffusion path as path analysis parameters, it effectively overcomes the technical problems of existing technologies that cannot accurately identify inspection blind spots after obstacle avoidance through the wave front diffusion algorithm and lack quantitative path accessibility analysis parameters. Thus, it achieves accurate positioning of potential inspection blind spots, obtains quantitative analysis parameters that can reflect the difficulty of path passage, and ensures the integrity of inspection coverage and the objectivity of path analysis.

[0038] like Figure 2 As shown, in another preferred embodiment of the present invention, based on path analysis parameters, blind spots in inspection caused by dynamic obstacles and corresponding monitoring points are automatically identified, and an anomaly risk assessment is initiated to obtain the anomaly risk assessment results, including: Spatial correlation matching was performed based on the potential inspection blind spots marked in the path analysis parameters and the dataset of missed monitoring points to confirm the actual inspection blind spots caused by dynamic obstacles and the associated monitoring points. Specifically, the following steps were taken: First, the potential inspection blind spot data marked in the path analysis parameters was retrieved. The data clearly identified three consecutive grids around the No. 2 water trough in the No. 3 livestock pen as potential blind spots, including the specific grid coordinate range. At the same time, the previously recorded dataset of missed monitoring points was extracted, which included the location information and monitoring items of the No. 2 water trough water level monitoring point. Correlation matching was performed by comparing spatial coordinates to check whether the grid range of each potential blind spot overlapped with the location of the missed monitoring point. It was found that the coverage area of ​​the No. 2 water level monitoring point exactly fell within the three grids of the potential blind spot. Thus, it was confirmed that the area was the actual inspection blind spot formed by the robot's obstacle avoidance deviation due to the gathering of multiple dairy cows, and the associated monitoring point was the No. 2 water level monitoring point. Other potential blind spots without overlapping missed monitoring points were ruled out, ensuring that the actual blind spot and the associated monitoring point corresponded accurately.

[0039] Based on the confirmed actual inspection blind spots, and combining the propulsion resistance coefficient in the path analysis parameters with the weights of key facility types associated with the blind spots, the baseline risk value of each blind spot is calculated. Specifically, this includes extracting the propulsion resistance coefficient data of the actual inspection blind spot and five adjacent passable grids. The grids are located in 1.2-meter narrow passages, 90-degree sharp turns, or areas with scattered feed. The coefficients are 2.0, 2.2, 2.1, 1.9, 2.3, 2.0, 1.8, and 2.2, respectively. The arithmetic mean of the average propulsion resistance coefficient is 2.1. Then, the weight rules for key facility types are retrieved: water troughs 3.0, ventilation equipment 2.5, and feed feeding points 2.0. Combining the passage difficulty reflected by the propulsion resistance coefficient with the monitoring importance reflected by the facility weights, the baseline risk value of the actual inspection blind spot is calculated through comprehensive weighting.

[0040] The baseline risk value is used to make a weighted correction by introducing a set of environmental impact factor weights, resulting in a weighted comprehensive risk value. Specifically, this includes: real-time collection of environmental data from the No. 3 livestock pen, determining that the environmental impact factor weight set includes three dimensions: dairy cow density, temperature and humidity, and ammonia concentration; at that time, there were three dairy cows staying around the blind spot, so the dairy cow density weight was set to 1.2; the real-time temperature of 28 degrees Celsius exceeded the suitable range for dairy cows, so the temperature and humidity weight was set to 1.1; and the ammonia concentration of 8 ppm was within the normal range, so the weight was set to 1.0; the comprehensive weight coefficient was calculated using the weighted average method to be 1.1, and then the baseline risk value was weighted and calculated with this coefficient to complete the correction of the baseline risk value.

[0041] Based on the weighted and corrected comprehensive risk value, a pre-defined risk level classification rule is used to determine the risk level, resulting in an anomaly risk assessment that includes the risk level, precise blind spot location, and a list of associated monitoring points. Specifically, this involves: invoking the pre-defined risk level classification rule, which is based on the conditions of a thousand-head dairy farm, industry standards, and management needs, clearly defining a comprehensive risk value below 5.0 as low risk, 5.0 to 8.0 as medium risk, and above 8.0 as high risk; and combining this with the corrected comprehensive risk value of 8.5, the blind spot is determined to be high risk. The data is then integrated to generate the assessment result.

[0042] In this embodiment of the invention, because the invention employs the following technical means: spatially matching the potential inspection blind spots marked in the path analysis parameters with the dataset of missed monitoring points to confirm the actual inspection blind spots and associated monitoring points; combining the propulsion resistance coefficient in the path analysis parameters with the weight of the key facility type associated with the blind spot to calculate the baseline risk value of each blind spot; introducing a weight set of environmental impact factors to weight and correct the baseline risk value to obtain a comprehensive risk value; and determining the level of the comprehensive risk value according to the preset risk level classification rules to obtain an abnormal risk assessment result containing the risk level, precise blind spot location, and a list of associated monitoring points, the invention overcomes the technical problems of existing technologies that cannot accurately distinguish between actual inspection blind spots caused by dynamic obstacles and associated monitoring points, lack the ability to quantitatively assess the risk of inspection blind spots, and are difficult to form risk assessment results with guiding significance. This achieves accurate positioning of actual inspection blind spots and associated monitoring points, scientific quantitative assessment of blind spot risks, and obtains complete and accurate abnormal risk assessment results, thereby improving the accuracy and effectiveness of livestock inspection risk management.

[0043] In a preferred embodiment of the present invention, when the abnormal risk assessment result indicates the existence of a potential hazard area, a graded early warning information is obtained, including the location of the potential hazard area, the expected missed monitoring points, and the risk level, comprising: The abnormal risk assessment results were analyzed to extract key data, including risk level, blind spot location, and a list of associated monitoring points. Specifically, key data were precisely extracted from the assessment results: the risk level was clearly high-risk; the blind spot location included detailed information such as livestock pen number 3, the precise coordinate range of each grid, and a coverage area of ​​0.75 square meters; the list of associated monitoring points included monitoring point identification numbers, monitoring point names (such as the No. 2 drinking trough water level monitoring point), specific monitoring items (water level and turbidity), and an effective monitoring coverage range of 1.5 meters; the data was categorized and organized, and redundant information was removed to ensure that the extracted key data was complete and accurate.

[0044] Based on the extracted key data and combined with predefined early warning level classification standards, the early warning level corresponding to each potential hazard area is determined. Specifically, this includes: retrieving the predefined early warning level classification standards, which are formulated based on the degree of impact of aquaculture risks and the urgency of the response. High risk corresponds to Level 1 early warning, requiring immediate initiation of review and inspection; medium risk corresponds to Level 2 early warning, requiring review to be arranged within 24 hours; and low risk corresponds to Level 3 early warning, which is included in the routine inspection supplement plan. Combining the key data extracted in the first step, the risk level of the current potential hazard area is confirmed as high risk, and the early warning level corresponding to this potential hazard area is directly matched and determined as Level 1 early warning according to the classification standards.

[0045] The warning levels are linked and integrated with the corresponding spatial coordinates of the potential hazard areas and the list of expected missed monitoring points to obtain a structured warning data set. Specifically, this includes: binding the determined Level 1 warning level with the corresponding spatial coordinates of the potential hazard areas, clarifying that the specific location indicated by the Level 1 warning is the spatial area and precise coordinates corresponding to the grid in the No. 3 livestock pen; at the same time, associating the warning level with the list of expected missed monitoring points, marking the missed monitoring point corresponding to the Level 1 warning as the No. 2 water trough water level monitoring point, including information such as monitoring items and coverage area; and integrating and sorting the three types of data—warning level, spatial coordinates of potential hazard areas, and list of expected missed monitoring points—according to a unified data structure standard to form a structured warning data set.

[0046] Based on the structured early warning data set, tiered early warning information is obtained, including the location of the potential hazard area, the expected missed monitoring points, and the risk level. Specifically, this includes: generating standardized tiered early warning information based on the structured early warning data set; clearly listing the core content, with the potential hazard area being the spatial area corresponding to the grid within the No. 3 livestock pen of the 1,000-head dairy farm, along with specific coordinates and coverage area; the expected missed monitoring point being the No. 2 water trough water level monitoring point, specifying the monitoring items as water level height and water turbidity; and the risk level being high risk, corresponding to a Level 1 early warning. The generated tiered early warning information is presented in a concise and intuitive format, allowing relevant personnel to quickly grasp the core information of the potential hazard, its priority, and its urgency.

[0047] In this embodiment of the invention, because the invention uses the technical means of extracting key data such as risk level, blind spot location and related monitoring point list from the abnormal risk assessment results, and combining them with predefined early warning level classification standards to determine the early warning level corresponding to each hidden danger area, and linking and integrating the early warning level with the spatial coordinates of the hidden danger area and the list of expected missed monitoring points to form a structured early warning data set, and then generating hierarchical early warning information based on this data set, it effectively overcomes the technical problem of existing technologies being unable to transform risk assessment results into clear, accurate and prioritized early warning information, which makes it difficult for keepers to quickly locate hidden danger areas, identify missed monitoring items and distinguish the priority of treatment. It realizes the structured presentation and hierarchical management of early warning information, allowing keepers to intuitively grasp the specific location of the hidden danger area, the monitoring points that need to be checked, and the urgency of the risk, providing clear guidance for quickly dispatching resources to carry out review inspections and improving the efficiency of abnormal response.

[0048] In a preferred embodiment of the present invention, based on tiered early warning information, the zookeeper is notified to reschedule the robot to re-enter the relevant area for a review and inspection, thereby completing the replanning of the route, including: This process involves analyzing tiered early warning information, extracting the spatial coordinates of potential hazard areas, risk level data, and a list of anticipated missed monitoring points. Based on the extracted spatial coordinates and risk level data, a priority evaluation algorithm is used to obtain a robot scheduling suggestion for ranking the review areas. Specifically, this includes: initiating the tiered early warning information analysis program; comprehensively decomposing the first-level tiered early warning information; accurately extracting three types of core data from the warning information; defining the spatial coordinates of potential hazard areas as the coordinate range corresponding to the grid within pen number three, including the precise coordinate values ​​of each grid; marking the risk level data as high risk, corresponding to the serious consequence of insufficient water intake for dairy cows, leading to a decrease in milk production; and detailing the list of anticipated missed monitoring points. The monitoring point identification number, monitoring point name (Water level monitoring point for No. 2 drinking trough), and core monitoring items (water level measurement and water turbidity detection) are listed in detail. The extracted spatial coordinate set of the potential hazard area and high-risk level data are used as core inputs to start the priority evaluation algorithm. The priority evaluation algorithm combines the breeding layout of the thousand-head dairy farm, the robot distribution status, and the current inspection task progress to comprehensively determine the factors with higher risk level weight. Since there is only one high-risk hazard area in this case, the algorithm directly lists the area as the highest review priority, generates a robot scheduling suggestion plan, and clearly suggests scheduling Robot No. 1, which is responsible for the daily inspection of No. 3 livestock pen, to go to the hazard area to perform the review inspection task first, and marks the scheduling priority as urgent.

[0049] The robot scheduling proposal is integrated with the list of anticipated missed monitoring points to construct a structured task instruction set containing inspection point sequences and review priorities. Specifically, this involves: deeply integrating the generated robot scheduling proposal with the list of anticipated missed monitoring points. The core instruction in the scheduling proposal is to prioritize scheduling robot #1 to the No. 3 livestock pen grid area. The list of anticipated missed monitoring points clearly defines the specific monitoring tasks to be completed. During the integration process, a structured task instruction set is constructed using review priorities as the core logic. The instruction set explicitly includes the inspection point sequence, meaning robot #1 must first arrive at the corresponding spatial location of the grid before moving sequentially to the livestock pen grid area. The review priority is marked as urgent, requiring the robot to initiate review within 10 minutes of receiving the instruction. Simultaneously, the monitoring tasks corresponding to each inspection point are detailed, namely, measuring the water level and detecting the turbidity of the No. 2 drinking trough, specifying that the measurement accuracy requirement is a water level error of no more than 0.5 cm, and the turbidity detection must meet industry safety standards.

[0050] Based on a structured task instruction set and combined with a real-time environmental grid map, the optimal review path sequence for the robot to reach each potential hazard area is planned and generated. Specifically, this includes: retrieving the real-time updated environmental grid map, which synchronizes the latest environmental status within livestock pen number three, including the current cow distribution location, passageway accessibility, and temporary obstacle information. Combining the emergency review requirements and inspection point sequence in the structured task instruction set, the path planning program is initiated. During the planning process, the current position of robot number 1, the specific coordinates of the potential hazard area, the cow aggregation situation in the real-time environment, and the passageway width are comprehensively considered. To avoid the robot encountering obstacles again and causing path deviation, an optimal review path sequence is planned, ensuring both passage efficiency and reducing collision risk.

[0051] The optimal verification path sequence and the corresponding structured task instruction set are sent to the target robot to perform the verification and inspection task. Specifically, the optimal verification path sequence and the corresponding structured task instruction set are simultaneously sent to the target robot, i.e., Robot No. 1. The sent optimal verification path sequence includes complete coordinate nodes, driving speed suggestions, and turning prompts. The structured task instruction set clarifies the monitoring items, measurement accuracy requirements, and data feedback frequency. At the same time, the sent instruction content is simultaneously pushed to the zookeeper's mobile terminal and the large display screen of the livestock pen monitoring center, informing the zookeeper that Robot No. 1 has accepted the emergency verification task, including task details, estimated arrival time, and real-time tracking method, to ensure that the zookeeper can monitor the robot's execution status throughout the process.

[0052] The system receives on-site verification data returned by the robot after performing the verification and inspection task. Based on the verification data, it verifies the status of potential hazard areas and updates the environmental database, completing dynamic optimization of path planning parameters. Specifically, this includes: Robot 1, following the issued path sequence and task instructions, completes the verification and inspection of the No. 3 livestock pen grid area, and then collects on-site verification data through sensors, including the actual water level of the No. 2 water trough, water turbidity detection values, and on-site environmental photos, and transmits this data back in real time; after receiving the verification data, it first verifies the status of potential hazard areas by comparing the actual water level with the preset standard water level to confirm whether there is a water shortage problem; it then compares the turbidity value with the safety threshold to determine whether the water quality meets the standards. If the data shows the water level... If the water quality meets the standards and is normal, the potential hazard is considered eliminated. If any abnormality is found, a secondary warning is immediately triggered. Subsequently, the verification data, hazard verification results, and on-site environmental photos are updated to the livestock pen environment database to supplement the latest monitoring records and dairy cow distribution dynamics for the area. At the same time, based on the actual situation of the obstacle avoidance path deviation and the verification path planning, the path planning parameters are dynamically optimized. The obstacle avoidance strategy, path adjustment threshold, and verification path priority weight when the robot encounters dynamic obstacles in the No. 3 livestock pen are adjusted to ensure that a more accurate and efficient inspection path can be planned when similar scenarios are encountered in the future. This forms a complete closed loop of warning, scheduling, verification, and optimization, solving the problem of monitoring omissions caused by path deviation that are difficult to remedy.

[0053] In this embodiment of the invention, because the invention uses the method of parsing hierarchical early warning information to extract the spatial coordinate set of hidden danger areas, risk level data, and a list of expected missed monitoring points, and generates a robot scheduling suggestion scheme for review area ranking through a priority evaluation algorithm, the scheduling scheme and the list of missed monitoring points are integrated to construct a structured task instruction set containing inspection point sequence and review priority, and the optimal review path sequence is planned in combination with a real-time environmental grid map and sent to the target robot, and the status of hidden dangers is verified after receiving review data, the environmental database is updated, and the path planning parameters are dynamically optimized, the invention overcomes the technical problems of existing technologies that lack a closed-loop review mechanism, lack of clear priority basis for the zookeeper to schedule robots, unscientific review path planning, and inability to optimize subsequent path parameters based on actual review results, resulting in the difficulty in effectively remedying monitoring omissions caused by path deviations. Thus, the invention achieves accurate scheduling, optimal path planning, and efficient execution of robot review and inspection, forming a complete closed loop of early warning, scheduling, review, and optimization, ensuring timely re-inspection of missed areas, improving the completeness, intelligence level, and refined management efficiency of livestock inspection, and further reducing the breeding risks caused by monitoring omissions.

[0054] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0055] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0056] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An anomaly early warning system based on intelligent path planning for livestock inspection robots, characterized in that: include: The monitoring module is used to collect environmental data in real time and monitor dynamic obstacles while the livestock inspection robot travels along the preset inspection path using the lidar and vision sensors configured on it. The recording module is used to immediately trigger the path replanning mechanism when a dynamic obstacle is detected that prevents the robot from traveling along the original path, obtain a new obstacle avoidance path, and record the preset monitoring points missed due to path deviation. The calculation module is used to construct a spatial topology based on the recorded path offset information and missed monitoring point data, using the midpoint of the water trough, the midpoint of the feed feeding point, and the midpoint of the ventilation equipment in the livestock pen, and obtain a grid map accordingly. On the grid map, wavefront diffusion is performed starting from the robot's original path and the new obstacle avoidance path to identify grid areas not covered by diffusion, and wavefront propulsion resistance is extracted as path analysis parameters. The assessment module is used to automatically identify blind spots and corresponding monitoring points caused by dynamic obstacles based on path analysis parameters, and to initiate anomaly risk assessment to obtain the anomaly risk assessment results. The acquisition module is used to obtain the location of the potential hazard area, the expected missed monitoring points, and the risk level classification warning information when the abnormal risk assessment results indicate that there is a potential hazard area. The verification module is used to notify the zookeeper to reschedule the robot to the relevant area for verification and inspection based on the graded early warning information, so as to complete the replanning of the route.

2. The anomaly early warning system based on intelligent path planning of livestock inspection robots according to claim 1, characterized in that, Equipped with lidar and vision sensors, the livestock inspection robot collects environmental data in real time and monitors dynamic obstacles while traveling along a preset inspection path, including: It receives LiDAR point cloud data and visual sensor image data uploaded by the robot, and performs spatiotemporal registration and fusion processing on the multi-source sensor data to obtain fused environmental perception data. Based on the fused environmental perception data, a real-time dynamic map of the robot's surrounding environment is obtained; the real-time dynamic map is then compared with a preset static environmental map to identify dynamic objects in the environment. The system continuously tracks identified dynamic objects and calculates their speed and direction. Based on the speed and direction of the dynamic objects, it predicts the future spatial intersection between the dynamic objects and the robot's preset inspection path. When a spatial intersection is predicted and the intersection time is less than a preset threshold, the system confirms that a dynamic obstacle has been detected.

3. The anomaly early warning system based on intelligent path planning of livestock inspection robots according to claim 2, characterized in that, When a dynamic obstacle is detected that prevents the robot from following its original path, a path replanning mechanism is immediately triggered to obtain a new obstacle avoidance path. Pre-defined monitoring points missed due to path deviation are also recorded, including: After confirming the detection of dynamic obstacles, the path planning algorithm is invoked. Starting from the robot's current position and with the goal of resuming the execution of the preset inspection task, a local path replanning is performed in the global grid map to obtain a new obstacle avoidance path. The new obstacle avoidance path is spatially compared with the preset inspection path, and the offset area between the two paths is calculated. Within the offset area, the preset monitoring points covered by the original preset inspection path but not covered by the new obstacle avoidance path are identified, and the preset monitoring points that are not covered are marked as missed monitoring points. The new obstacle avoidance path, path offset information, and the missing monitoring point data are associated to generate a structured path offset record.

4. The anomaly early warning system based on intelligent path planning of livestock inspection robots according to claim 3, characterized in that, Based on recorded path offset information and missed monitoring point data, a spatial topology is constructed using the midpoints of the water troughs, feed feeding points, and ventilation equipment within the livestock pen, and a raster map is obtained accordingly, including: Based on the geospatial information in the path offset records, the spatial coordinates of the midpoints of all water troughs, feed feeding points and ventilation equipment are extracted from the predefined livestock pen structure layout database. By extracting the midpoint coordinates of each facility as topological nodes, and determining the connection relationship between nodes based on the actual passage layout within the livestock pen, a topological network representing the spatial connectivity between facilities is constructed. The constructed topology network is used as input and projected onto a two-dimensional Cartesian coordinate plane. The plane is then divided into grids using a preset resolution to obtain a two-dimensional raster map with topological constraints. The geographic coordinates of the original preset inspection path and the new obstacle avoidance path in the path offset record are mapped onto a two-dimensional raster map with topological constraints to obtain a path label map containing two path raster sequences.

5. The intelligent path planning anomaly early warning system based on livestock inspection robots according to claim 4, characterized in that, On the grid map, wavefront diffusion is performed starting from the robot's original path and the new obstacle avoidance path to identify grid areas not covered by the diffusion, and wavefront propulsion resistance is extracted as path analysis parameters, including: Based on the path-marked map, the grid sequence of the original preset inspection path and the new obstacle avoidance path is initialized as the starting wave source for wavefront diffusion. The wavefront diffusion algorithm is executed starting from the initial wave source. During the diffusion process, the wavefront arrival value of each grid is recorded, and the corresponding propulsion drag coefficient is calculated based on the channel width, turning angle and obstacle distribution of the grid. Based on the relationship between wavefront arrival value and a preset threshold, grid areas whose arrival value exceeds the threshold and are not covered by the wavefront are identified and marked as potential inspection blind spots. Based on the propulsion drag coefficient and wavefront diffusion path, the average drag intensity and maximum drag value in each diffusion direction are extracted to obtain path analysis parameters for evaluating path reachability.

6. The anomaly early warning system based on intelligent path planning of livestock inspection robots according to claim 5, characterized in that, Based on path analysis parameters, the system automatically identifies blind spots in inspections caused by dynamic obstacles and corresponding monitoring points, and initiates anomaly risk assessment to obtain the assessment results, including: Spatial correlation matching is performed on the potential inspection blind spots marked in the path analysis parameters and the dataset of missed monitoring points to identify the actual inspection blind spots caused by dynamic obstacles and the associated monitoring points. Based on the confirmed actual inspection blind spots, the baseline risk value of each blind spot is calculated by combining the propulsion resistance coefficient in the path analysis parameters and the weight of the key facility type associated with the blind spot. The baseline risk value is used to make a weighted correction by introducing a set of environmental impact factor weights, resulting in a weighted comprehensive risk value. Based on the weighted and corrected comprehensive risk value, the risk level is determined using the preset risk level classification rules, resulting in an abnormal risk assessment result that includes the risk level, precise blind spot location, and a list of associated monitoring points.

7. The anomaly early warning system based on intelligent path planning of livestock inspection robots according to claim 6, characterized in that, When the abnormal risk assessment results indicate the existence of a potential hazard area, the system provides graded early warning information including the location of the hazard area, the expected missed monitoring points, and the risk level, including: Analyze the abnormal risk assessment results and extract key data such as risk level, blind spot location, and list of associated monitoring points; Based on the extracted key data and combined with the predefined early warning level classification standards, the early warning level corresponding to each potential hazard area is determined; By associating and integrating the warning level with the corresponding spatial coordinates of the hidden danger area and the list of expected missed monitoring points, a structured set of warning data is obtained; Based on the structured early warning data set, graded early warning information is obtained, including the location of potential hazard areas, the expected missed monitoring points, and the risk level.

8. The intelligent path planning anomaly early warning system based on livestock inspection robots according to claim 7, characterized in that, Based on the tiered early warning information, the zookeeper is notified to reschedule the robot to the relevant area for a re-inspection, in order to complete the replanning of the route, including: The hierarchical early warning information is analyzed, and the spatial coordinate set of the hidden danger area, the risk level data, and the list of expected missed monitoring points are extracted from the hierarchical early warning information. Based on the extracted spatial coordinate set of the hidden danger area and the risk level data, a robot scheduling suggestion scheme for the review area ranking is obtained through a priority evaluation algorithm. The robot scheduling suggestion scheme is integrated with the list of expected missed monitoring points to construct a structured task instruction set that includes the inspection point sequence and review priority; Based on the structured task instruction set and combined with the real-time environmental grid map, the optimal review path sequence for the robot to go to each potential hazard area is planned and generated. The optimal review path sequence and the corresponding structured task instruction set are sent to the target robot to perform the review and inspection task; The system receives on-site verification data returned by the robot after it performs the verification and inspection task. Based on the verification data, it verifies the status of the hidden danger area and updates the environmental database, thus completing the dynamic optimization of the path planning parameters.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to perform the system as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, performs the system as described in any one of claims 1 to 8.

Images