Method, device and equipment for identifying loading abnormality based on density feature ratio, and method, device, equipment, medium and system for checking

Abstract

The application provides a loading anomaly identification method based on a density feature ratio, a checking method, device, electronic equipment, computer readable storage medium and system. The method performs coordinate axis calibration on a vehicle compartment point cloud sequence, constructs vehicle compartment cavity loading space voxel data of a vehicle to be inspected, divides the vehicle compartment cavity loading space voxel data into multiple spatial grid units, counts local filling density values in each spatial grid unit to generate a real-time loading density distribution atlas of the vehicle to be inspected, converts the real-time loading density distribution atlas into a loading density feature ratio vector, compares the loading density feature ratio vector with standard density distribution feature data of a corresponding agricultural product category in a three-dimensional grid, classifies the loading state of the vehicle to be inspected, determines whether the vehicle to be inspected has an internal loading anomaly, and generates an anomaly determination result. When the anomaly determination result indicates that there is an internal loading anomaly, a visualized text review instruction is generated based on the anomaly determination result.

Description

Technical Field

[0001] This application relates to the field of vehicle passage inspection technology, and more specifically, to a method, device, electronic device, computer-readable storage medium and system for identifying and inspecting loading anomalies based on density feature ratio. Background Technology

[0002] In the inspection of green channel vehicles on highways, the vehicles are characterized by a wide variety of cargo, significant differences in loading methods, and marked variations in cargo compartment structure. The inspection system needs to quickly identify the vehicle's loading status during transit and link the identification results with the lane inspection equipment. As highway toll station inspection processes gradually move towards digitalization, how to obtain data reflecting the vehicle's loading status without affecting traffic efficiency, and how to technically assess the match between the declared cargo categories and the actual loading status, have become crucial technical requirements for identifying anomalies in green channel vehicles.

[0003] Existing green channel vehicle inspection schemes typically rely on lane weighing data, vehicle exterior dimensions, and declared product category data as the primary basis for judgment. This scheme first collects vehicle weight and exterior dimensions as the vehicle passes through the inspection lane, then compares the collected results with the declared information and threshold rules, and finally determines whether the vehicle requires further verification based on the comparison results. The key focus of this scheme is judging the macroscopic consistency between the overall vehicle weight, vehicle model exterior dimensions, and declared information.

[0004] However, the above-mentioned solution still has certain technical shortcomings. Because it primarily relies on overall indicators such as total weight and external dimensions, it struggles to reflect variations in loading density between different spatial locations within the vehicle compartment. Furthermore, it is difficult to establish a positional correspondence between the current vehicle loading status and the standard loading characteristics of the corresponding agricultural product category in three-dimensional space. When the vehicle exhibits loading anomalies such as partial vacancy, inclusion of non-agricultural products, foreign objects at the bottom, or layered mixed loading, relying solely on overall weight or external dimensions can easily mask these localized spatial differences with overall indicators, resulting in low specificity in identifying loading anomalies within green channel vehicles. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, computer-readable storage medium, and system for identifying and inspecting loading anomalies based on density feature ratio, so as to at least alleviate the technical problems of existing green channel vehicle inspection schemes that are difficult to reflect the changes in loading density between different spatial locations inside the vehicle compartment and difficult to establish a three-dimensional spatial correspondence between the current vehicle loading status and the standard loading characteristics of the corresponding agricultural product category.

[0006] A method for identifying and verifying loading anomalies based on density feature ratio, comprising: In response to the loading inspection trigger command of the carriage to be inspected, the point cloud sequence of the carriage to be inspected is obtained; The coordinate axis of the carriage point cloud sequence is calibrated, and the body shell feature data and wheel feature data in the carriage point cloud sequence are removed to construct the carriage interior loading space voxel data of the carriage to be inspected. The voxel data of the loading space inside the carriage is divided into multiple spatial grid cells. The local filling density value in each spatial grid cell is counted. The regional density change value is determined based on the difference in the local filling density value between adjacent spatial grid cells to generate a real-time loading density distribution map of the carriage to be inspected. The real-time loading density distribution map is then converted into a loading density feature ratio vector. The loading density feature vector is compared with the standard density distribution feature data of the corresponding agricultural product category in a three-dimensional grid. Based on the spatial density difference generated by the comparison, the loading status of the inspection compartment is classified into patterns to determine whether there is an internal loading abnormality in the inspection compartment, so as to generate an abnormality judgment result. When the anomaly determination result indicates the existence of an internal loading anomaly, a visual text verification instruction is generated based on the anomaly determination result, and the text verification instruction is pushed to an external verification terminal for manual verification of the anomaly.

[0007] A device for identifying and inspecting loading anomalies based on density feature ratio, comprising: The carriage point cloud sequence acquisition module is configured to acquire the carriage point cloud sequence of the carriage to be inspected in response to the loading inspection trigger command of the carriage to be inspected; The voxel data construction module is configured to perform coordinate axis calibration on the point cloud sequence of the carriage and remove the body shell feature data and wheel feature data from the point cloud sequence of the carriage to construct the voxel data of the loading space inside the carriage cavity of the carriage to be inspected. The density feature ratio vector generation module is configured to divide the voxel data of the loading space inside the carriage into multiple spatial grid cells, count the local filling density value in each spatial grid cell, and determine the regional density change value based on the difference in the local filling density value between adjacent spatial grid cells, so as to generate a real-time loading density distribution map of the carriage to be inspected, and convert the real-time loading density distribution map into a loading density feature ratio vector. The anomaly determination result generation module is configured to perform a three-dimensional grid comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, and to perform pattern classification on the loading status of the carriage to be inspected based on the spatial density difference generated by the comparison, thereby determining whether there is an internal loading anomaly in the carriage to be inspected, so as to generate an anomaly determination result. The text verification instruction push module is configured to generate a visual text verification instruction based on the anomaly determination result when the anomaly determination result indicates that there is an internal loading anomaly, and push the text verification instruction to an external verification terminal for manual verification of the anomaly.

[0008] An electronic device includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the loading anomaly identification and verification method based on density feature ratio as described in any one of the claims.

[0009] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the loading anomaly identification and verification method based on density feature ratio as described in any one of the claims.

[0010] A loading anomaly identification and inspection system based on density feature ratio includes a point cloud acquisition device, an anomaly pattern classification and identification device, and an external inspection terminal. The point cloud acquisition device is used to respond to the loading inspection trigger command of the carriage to be inspected, acquire the carriage point cloud sequence of the carriage to be inspected, and send the carriage point cloud sequence to the abnormal pattern classification and recognition device. The abnormal pattern classification and recognition device is used to perform coordinate axis calibration on the carriage point cloud sequence and remove the body shell feature data and wheel feature data from the carriage point cloud sequence in order to construct the carriage interior loading space voxel data of the carriage to be inspected. The abnormal pattern classification and recognition device is also used to divide the voxel data of the loading space inside the carriage into multiple spatial grid units, count the local filling density value in each spatial grid unit, and determine the regional density change value based on the difference in the local filling density value between adjacent spatial grid units, so as to generate a real-time loading density distribution map of the carriage to be inspected, and convert the real-time loading density distribution map into a loading density feature ratio vector. The abnormal pattern classification and recognition device is also used to perform a three-dimensional grid comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, and to perform pattern classification on the loading status of the carriage to be inspected based on the spatial density difference generated by the comparison, to determine whether there is an internal loading abnormality in the carriage to be inspected, so as to generate an abnormality judgment result. The abnormal pattern classification and recognition device is also used to generate a visual text verification instruction based on the abnormal judgment result when the abnormal judgment result indicates that there is an internal loading abnormality, and push the text verification instruction to the external verification terminal. The external verification terminal is used to receive the text verification instruction and to visually display the text verification instruction.

[0011] The technical advantages of the technical solution provided in this application are: This application presents a method for identifying and inspecting loading anomalies based on density feature ratio. Addressing the shortcomings of existing green channel loading inspection schemes, which primarily rely on overall indicators such as total weight and external dimensions, making it difficult to reflect variations in loading density across different spatial locations within the vehicle compartment, this method first responds to the loading inspection trigger command of the vehicle to be inspected and acquires the vehicle compartment's point cloud sequence. This allows subsequent identification processing to move beyond focusing on the vehicle's overall weight or external dimensions, instead using the point cloud sequence, which reflects the internal spatial state of the compartment, as the processing basis. Compared to traditional methods that rely on macroscopic indicators, this application acquires the point cloud sequence of the vehicle compartment at the moment of loading inspection triggering, providing a more direct data source for subsequent characterization of the loading space within the compartment. This helps reduce the possibility of overall indicators masking local loading differences.

[0012] Furthermore, this application performs coordinate axis calibration on the point cloud sequence of the carriage and removes body shell feature data and wheel feature data from the point cloud sequence to construct voxel data of the loading space inside the carriage to be inspected. This processing addresses the technical problem of existing solutions' inability to distinguish between the external structure and the internal loading area of ​​the carriage. It removes point cloud features related to external physical components such as the body shell and wheels from the subsequent loading analysis objects, allowing subsequent density analysis to focus on the loading space inside the carriage. Compared to using the overall scan results or outline data directly as the basis for judgment, the voxel data of the loading space inside the carriage more closely reflects the internal loading state of the carriage to be inspected, providing a more targeted spatial data foundation for subsequent spatial grid cell division, density change extraction, and loading anomaly pattern classification.

[0013] Furthermore, this application divides the voxel data of the loading space inside the carriage into multiple spatial grid cells, statistically analyzes the local filling density values ​​within each spatial grid cell, and determines the regional density change value based on the difference in local filling density values ​​between adjacent spatial grid cells to generate a real-time loading density distribution map of the carriage under inspection. This real-time loading density distribution map is then converted into a loading density feature ratio vector. This processing addresses the shortcomings of existing solutions in expressing local spatial density differences and forming classifiable feature representations. It transforms the loading space inside the carriage from a holistic object into multiple spatial grid cells with spatial relationships, and further converts the local filling density values, regional density change values, and their spatial relationships into a loading density feature ratio vector. Compared to methods that only generate maps or only perform overall threshold judgments, this application can convert the density distribution differences in different areas inside the carriage into data representations that can participate in subsequent comparisons and pattern classifications, giving a clearer feature representation basis for abnormal morphologies such as local vacancy, local inclusions, bottom-mounted foreign objects, or layered loading.

[0014] Furthermore, this application performs a three-dimensional mesh comparison between the loading density feature vector and the standard density distribution feature data of the corresponding agricultural product category. Based on the spatial density differences generated by the comparison, it performs pattern classification on the loading status of the inspection compartment to determine whether there are internal loading anomalies in the inspection compartment, thereby generating an anomaly determination result. This process addresses the technical shortcomings of existing solutions, which struggle to establish a three-dimensional spatial correspondence between the current loading status and the standard loading features of the corresponding agricultural product category, and whose anomaly determination methods tend to rely on macroscopic comparisons. By introducing the loading density feature vector and the standard density distribution feature data of the corresponding agricultural product category into the same comparison process, the anomaly determination result is no longer determined solely by the total threshold or outline information, but rather by the spatial density differences generated by the three-dimensional mesh comparison and the pattern classification result of the loading status. Compared to the traditional approach of macroscopically comparing declared information and overall indicators, this application can identify the spatial differences between the internal loading status of the inspection compartment and the standard distribution under the density feature caliber of the corresponding agricultural product category. This makes the determination of internal loading anomalies closer to the actual spatial manifestation of local loading anomalies within the compartment and improves the specificity of the anomaly identification results.

[0015] Furthermore, when the anomaly determination result indicates the existence of an internal loading anomaly, this application generates a visual text verification instruction based on the anomaly determination result and pushes the text verification instruction to an external inspection terminal for manual verification of the anomaly. This process addresses the problem of insufficient linkage between anomaly identification results and on-site verification terminals in existing solutions. It transforms the anomaly determination result obtained based on loading density feature ratio vectors, spatial density differences, and pattern classification into a visual text verification instruction, which is then presented through an external inspection terminal. Compared to processing methods that only generate background judgment results, this application enables a direct information transmission link between the internal loading anomaly identification result and the manual inspection process. This allows inspectors to locate the anomaly area to be verified based on the text verification instruction, thereby improving the usability of the internal loading anomaly identification result in on-site inspection. Attached Figure Description

[0016] Figure 1 This application provides an embodiment of a scenario for identifying and verifying loading anomalies based on density feature ratio. Figure 2 This application provides a method for identifying and verifying loading anomalies based on density feature ratio. Figure 3 This application provides an embodiment of a device for identifying and inspecting loading anomalies based on density feature ratio. Figure 4 This is an electronic device according to an embodiment of the present application.

[0017] Figure 5 This is a computer-readable storage medium according to an embodiment of the present application.

[0018] Figure 6 This application provides an embodiment of a loading anomaly identification and verification system based on density feature ratio. Detailed Implementation

[0019] like Figure 1 As shown, this is an embodiment of the present application of a scenario for identifying and verifying loading anomalies based on density feature ratio; as Figure 2 The image shows an embodiment of this application of a method for identifying and verifying loading anomalies based on density feature ratio, comprising the following steps: In response to the loading inspection trigger command of the carriage to be inspected, the point cloud sequence of the carriage to be inspected is obtained; The coordinate axis of the carriage point cloud sequence is calibrated, and the body shell feature data and wheel feature data in the carriage point cloud sequence are removed to construct the carriage interior loading space voxel data of the carriage to be inspected. The voxel data of the loading space inside the carriage is divided into multiple spatial grid cells. The local filling density value in each spatial grid cell is counted. The regional density change value is determined based on the difference in the local filling density value between adjacent spatial grid cells to generate a real-time loading density distribution map of the carriage to be inspected. The real-time loading density distribution map is then converted into a loading density feature ratio vector. The loading density feature vector is compared with the standard density distribution feature data of the corresponding agricultural product category in a three-dimensional grid. Based on the spatial density difference generated by the comparison, the loading status of the inspection compartment is classified into patterns to determine whether there is an internal loading abnormality in the inspection compartment, so as to generate an abnormality judgment result. When the anomaly determination result indicates the existence of an internal loading anomaly, a visual text verification instruction is generated based on the anomaly determination result, and the text verification instruction is pushed to an external verification terminal for manual verification of the anomaly.

[0020] The above Figure 2 The proposed method revolves around the point cloud sequence of the freight cars under inspection at the loading and inspection station. First, the point cloud sequence is converted into voxel data of the loading space within the freight car's interior. Then, this voxel data is converted into a real-time loading density distribution map that expresses the loading density relationship at different spatial locations. Further, the real-time loading density distribution map is converted into a loading density feature ratio vector. Subsequently, the loading density feature ratio vector is compared with the standard density distribution feature data of the corresponding agricultural product category using the same grid dimension in a three-dimensional grid. Based on the spatial density differences generated by the comparison, the loading status of the freight cars under inspection is classified into patterns. This ensures that the anomaly determination results are based on the internal spatial location and feature expression of the freight car, rather than solely relying on overall weight, external dimensions, or manual visual judgment.

[0021] Specifically, in this application, the carriage to be inspected is the carriage object that enters the loading and inspection station and requires internal loading status identification. The loading and inspection trigger command can be formed by converting one of the following signals: carriage arrival detection signal, scan preparation completion signal, inspection task creation signal, or carriage positioning completion signal. The loading and inspection trigger command is not a simple prompt message, but rather a set of start data including the trigger time, inspection station identifier, carriage object identifier, and scan start status after the carriage to be inspected enters the loading and inspection station. When responding to the loading and inspection trigger command of the carriage to be inspected, the point cloud frame corresponding to the trigger time in the loading and inspection trigger command can be read from the carriage scanning device's cache, starting from the trigger time in the loading and inspection trigger command, and then arranged into a carriage point cloud sequence according to the acquisition order of the point cloud frames. Each point in the carriage point cloud sequence can include fields such as horizontal coordinates, vertical coordinates, acquisition time, and reflection intensity. The horizontal coordinates represent the point's position along the carriage width, the vertical coordinates along the carriage length, and the vertical coordinates along the carriage height. The acquisition time is used to sequentially connect different point cloud frames, and the reflection intensity helps distinguish between the surface of the load and the surface of external physical components. The carriage point cloud sequence formed in this way can cover the spatial reflection information of both the internal loading volume of the carriage and the external physical components. Subsequent coordinate axis calibration of the carriage point cloud sequence can use this sequence as a unified data source.

[0022] Specifically, in this application, the point cloud sequence of the carriage is typically acquired using the installation position of the carriage scanning equipment or the scanning reference position of the loading and inspection station as coordinate references. Since the carriage to be inspected may experience lateral offset, longitudinal deflection, or the carriage centerline not being completely aligned with the scanning reference line when placed or parked at the loading and inspection station, directly calculating the point cloud distribution in the carriage point cloud sequence under the original coordinates would cause the meaning of the subsequent spatial grid units in the lateral, longitudinal, and vertical axes to be inconsistent with the actual carriage orientation. Therefore, when calibrating the coordinate axes of the carriage point cloud sequence, point cloud points that reflect the overall extension direction of the carriage to be inspected can be extracted first, and then the principal component directions can be obtained based on the discrete directions of these point cloud points in space. The principal component direction can be understood as the direction of the greatest change in the point cloud distribution in the carriage point cloud sequence, which has a strong correspondence with the longitudinal extension direction of the carriage to be inspected from one end to the other. After taking the principal component direction as the longitudinal centerline of the carriage to be inspected, the carriage point cloud sequence in the original coordinate system is rotated and translated into a standardized coordinate system based on the longitudinal centerline of the carriage. This ensures that the longitudinal axis of the standardized coordinate system corresponds to the length direction of the carriage, the lateral axis corresponds to the width direction, and the vertical axis corresponds to the height direction. The carriage point cloud sequence, after coordinate axis calibration, allows the placement deviations of different carriages at the loading and inspection station to be expressed under a unified carriage orientation, thus providing a stable spatial reference for subsequent removal of body shell feature data and wheel feature data.

[0023] Specifically, in this application, the step of performing coordinate axis calibration on the carriage point cloud sequence and removing the body shell feature data and wheel feature data from the carriage point cloud sequence to construct the carriage interior loading space voxel data of the carriage to be inspected includes: establishing the initial three-dimensional coordinate values ​​of the carriage point cloud sequence in a spatial rectangular coordinate system; determining the longitudinal centerline of the carriage to be inspected using the principal component directions of the initial three-dimensional coordinate values, and converting the initial three-dimensional coordinate values ​​to a standardized coordinate system based on the longitudinal centerline of the carriage to complete the coordinate axis calibration of the carriage point cloud sequence; identifying point cloud points belonging to external physical components of the carriage in the carriage point cloud sequence according to a preset spatial boundary contour threshold, and determining the body shell feature data and wheel feature data based on the point cloud points of the external physical components of the carriage, so as to remove the body shell feature data and wheel feature data from the coordinate axis calibrated carriage point cloud sequence, and constructing the carriage interior loading space voxel data based on the retained carriage interior loading volume space of the carriage to be inspected.

[0024] Specifically, in this application, the initial three-dimensional coordinate values ​​are the positional representations of point cloud points in the carriage point cloud sequence within the original spatial rectangular coordinate system. The lateral, longitudinal, and vertical coordinates in the initial three-dimensional coordinate values ​​are derived from the spatial ranging results of the reflection positions on the surface of the carriage to be inspected by the carriage scanning device. When establishing the initial three-dimensional coordinate values ​​of the carriage point cloud sequence in the spatial rectangular coordinate system, the original ranging fields of each point cloud point in the carriage point cloud sequence can be read first, and then the original ranging fields can be converted into initial three-dimensional coordinate values ​​in the spatial rectangular coordinate system according to the installation orientation of the carriage scanning device. After the initial three-dimensional coordinate values ​​are established, the center position of the point cloud corresponding to the initial three-dimensional coordinate values ​​can be calculated first, and the offset of each point cloud point relative to the center position of the point cloud can be used as direction analysis data. When determining the principal component directions based on the direction analysis data, the discrete relationship between the lateral, longitudinal, and vertical coordinates can be calculated, and the direction with a higher degree of point cloud point extension can be selected from the discrete relationship as the principal component direction. The principal component direction is used to determine the longitudinal centerline of the carriage to be inspected. After the longitudinal centerline is determined, the initial three-dimensional coordinate values ​​can be rotated according to the longitudinal centerline, and the stable point cloud position near the end of the carriage or the entrance of the carriage cavity is used as a translation reference to transform to a standardized coordinate system. After the standardized coordinate system is formed, each point in the carriage point cloud sequence has a spatial position corresponding to the direction of the carriage itself, so that the subsequent spatial boundary contour threshold can be used under a unified coordinate system.

[0025] Specifically, in this application, the preset spatial boundary contour threshold is used to describe the spatial boundary conditions between the external physical components of the vehicle body and the internal loading volume space of the vehicle body to be inspected. The preset spatial boundary contour threshold can be pre-configured after the loading inspection station is deployed using empty vehicle body scanning data, vehicle body outline calibration data, and internal loading volume space calibration data. The preset spatial boundary contour threshold may include the lateral boundary range of the outer surface of the vehicle body sidewall, the vertical boundary range of the outer surface of the vehicle body top, the vertical rejection range near the bottom of the vehicle body, and the arc-shaped outer edge range of the wheel area. When identifying the point cloud points of the external physical components of the vehicle body according to the preset spatial boundary contour threshold, the position of the vehicle body point cloud sequence after coordinate axis calibration can be compared with the preset spatial boundary contour threshold. Point cloud points located near the lateral boundary range of the outer surface of the vehicle body sidewall and continuously distributed along the longitudinal direction can be identified as the source point cloud points of the vehicle body shell feature data, and point cloud points located near the bottom of the vehicle body and exhibiting a local arc distribution in the longitudinal direction can be identified as the source point cloud points of the wheel feature data. The vehicle body shell feature data is used to express the spatial response of the vehicle body shell, the outer surface of the side walls, and the outer surface of the roof in the vehicle body point cloud sequence. The wheel feature data is used to express the spatial response of the wheels and tire edges in the vehicle body point cloud sequence. After removing the vehicle body shell feature data and wheel feature data from the vehicle body point cloud sequence after coordinate axis calibration, the remaining point cloud points are concentrated within the loading volume space inside the vehicle body to be inspected. Subsequent voxelization processing will no longer include the vehicle body shell feature data and wheel feature data as load data in the statistics.

[0026] Specifically, in this application, the voxel data of the loading space inside the carriage is a data representation of the loading volume space inside the carriage to be inspected, discretized according to three-dimensional spatial units. When constructing the voxel data of the loading space inside the carriage, the lateral, longitudinal, and vertical boundaries of the loading volume space inside the carriage to be inspected can be determined first based on the retained loading volume space inside the carriage to be inspected. Then, according to the spatial scale required for subsequent spatial grid unit division, the loading volume space inside the carriage to be inspected is decomposed into continuously arranged voxel positions. Each voxel position can record its corresponding lateral coordinate interval, longitudinal coordinate interval, vertical coordinate interval, and the number of point cloud points falling into that voxel position. The number of point cloud points falling into that voxel position is used to express the degree to which that voxel position is occupied by the surface of agricultural products or objects inside the carriage. The lateral, longitudinal, and vertical coordinate intervals are used to enable that voxel position to participate in the positioning and adjacency determination of subsequent spatial grid units. Since the voxel data of the loading space inside the carriage has removed the feature data of the vehicle body shell and the wheel feature data, the subsequent local filling density value, regional density change value and loading density feature ratio vector can more centrally reflect the loading status inside the carriage, without being interfered with by the point cloud response of the physical components outside the carriage.

[0027] Specifically, in this application, the step of dividing the voxel data of the loading space inside the carriage cavity into multiple spatial grid cells, statistically analyzing the local filling density values ​​within each spatial grid cell, and determining the regional density change value based on the difference in local filling density values ​​between adjacent spatial grid cells to generate a real-time loading density distribution map of the carriage to be inspected, and converting the real-time loading density distribution map into a loading density feature ratio vector, includes: dividing the voxel data of the loading space inside the carriage cavity into voxels along the horizontal, vertical, and longitudinal axes according to a preset grid step size to form multiple cuboid-shaped spatial grid cells; statistically analyzing the reflections contained within each spatial grid cell. The number of point clouds is determined, and the local filling density value of the corresponding spatial grid cell is determined based on the number of reflection point clouds. The change difference between the local filling density values ​​of two adjacent spatial grid cells is calculated, and the regional density change value of the corresponding spatial grid cell is determined based on the change difference. The real-time loading density distribution map is generated based on the regional density change values ​​corresponding to multiple spatial grid cells. According to the arrangement order of the spatial grid cells in the horizontal, vertical and vertical axes, the local filling density value and regional density change value corresponding to each spatial grid cell are positionally bound to convert the real-time loading density distribution map into the loading density feature ratio vector.

[0028] Specifically, in this application, the preset grid step size is used to define the spatial unit when the voxel data of the loading space inside the carriage is divided. The preset grid step size can be configured according to the point cloud sampling interval of the carriage scanning equipment, the particle scale of the loading inside the carriage, and the spatial positioning granularity required by the external inspection terminal. For agricultural products such as apples, oranges, and potatoes, the preset grid step size can be configured to be larger than the size of the shape fluctuation of a single agricultural product and smaller than the spatial size of local loading anomalies in the carriage, so that the spatial grid unit is not excessively affected by local protrusions on the surface of a single agricultural product, and can also preserve spatial differences such as empty front of the carriage, crammed in the middle of the carriage, or abnormal padding at the bottom of the carriage. When voxelizing according to the preset grid step size, the grid boundaries can be determined in the horizontal, vertical, and longitudinal axes respectively, and the grid boundaries in the three axes are combined into multiple cuboid-shaped spatial grid units. The horizontal range of the spatial grid unit corresponds to a position segment in the width direction of the carriage, the longitudinal range of the spatial grid unit corresponds to a position segment in the length direction of the carriage, and the vertical range of the spatial grid unit corresponds to a position segment in the height direction of the carriage. After the spatial grid cells are formed, each spatial grid cell has an inclusion relationship with several voxel positions in the voxel data of the loading space inside the carriage cavity. The loading space voxel data of the loading space inside the carriage cavity is thus converted into a three-dimensional spatial cell sequence that can perform local density statistics and feature ratio expression.

[0029] Specifically, in this application, the number of reflected point clouds is the count of point cloud points falling within the same spatial grid cell. This count originates from the retained point cloud points after the carriage point cloud sequence has undergone coordinate axis calibration, removal of body shell feature data and wheel feature data, and construction of the carriage interior loading space voxel data. When counting the number of reflected point clouds contained in each spatial grid cell, the horizontal, vertical, and lateral coordinates of each retained point cloud point can be compared with the horizontal, vertical, and lateral ranges of the spatial grid cell, respectively. When a retained point cloud point falls within all three coordinate ranges of the same spatial grid cell, the retained point cloud point is included in the number of reflected point clouds of the corresponding spatial grid cell. When determining the local fill density value of the corresponding spatial grid cell based on the number of reflected point clouds, if all spatial grid cells have the same volume, the number of reflected point clouds can be normalized after being processed by the number of acquisition frames to obtain the local fill density value. If a non-complete cuboid spatial grid cell is formed at the edge of the loading space voxel data within the carriage cavity, the number of reflected point clouds can be normalized according to the actual volume of the spatial grid cell to obtain the local fill density value. The local fill density value describes the degree of reflection and coverage within the corresponding spatial grid cell by the loaded object or objects inside the carriage. After the local fill density value is used in the calculation of the difference in variation between adjacent spatial grid cells, it can express the variation in loading density between different locations inside the carriage.

[0030] Specifically, in this application, adjacent spatial grid cells refer to spatial grid cells that have adjacent coordinate boundaries in the horizontal, vertical, or transverse directions. When calculating the change difference based on the local fill density values ​​between two adjacent spatial grid cells, the current spatial grid cell can be selected first, and then the local fill density values ​​of the spatial grid cells adjacent to the current spatial grid cell in the horizontal, vertical, and transverse directions can be read. For each pair of adjacent spatial grid cells, the difference between the two local fill density values ​​can be used as the change difference; when it is necessary to express the direction of density change, the density rise and fall relationship from the current spatial grid cell to the adjacent spatial grid cell can also be retained. When determining the regional density change value of the corresponding spatial grid cell based on the change difference, the change difference of the current spatial grid cell in different adjacent directions can be non-negatively processed, and then synthesized according to a preset direction priority to obtain the regional density change value corresponding to the spatial grid cell. The preset direction priority can be configured according to the spatial directions in which anomalies are more likely to occur inside the carriage, for example, assigning higher participation weights to longitudinal continuous vacancy and vertical layering changes. The regional density change value is not simply a point cloud count, but spatial change data used to express the degree of abrupt change in the fill state between the current spatial grid cell and adjacent spatial grid cells. By combining the regional density change values ​​corresponding to multiple spatial grid cells according to the arrangement relationship of the spatial grid cells in the loading space voxel data of the carriage cavity, the real-time loading density distribution map can be generated.

[0031] Specifically, in this application, the real-time loading density distribution map is a three-dimensional data map corresponding to the current loading inspection status of the carriage to be inspected. Each map node in the real-time loading density distribution map corresponds to a spatial grid cell. The map node records the horizontal coordinate range, vertical coordinate range, and vertical coordinate range of the spatial grid cell, as well as the local fill density value and the regional density change value. In terms of data processing, the real-time loading density distribution map is not limited to an image for display, but is used as the data basis for generating loading density feature ratio vectors, performing three-dimensional grid comparisons, and forming internal loading anomaly judgment results. If it is necessary to present the real-time loading density distribution map on the external inspection terminal, the regional density change value can be converted into grayscale depth or color level; however, in the anomaly judgment process of this application, the spatial grid cell coordinates, local fill density values, and regional density change values ​​recorded in the real-time loading density distribution map are still used as the calculation basis. Because the real-time loading density distribution map preserves the spatial relationship inside the carriage, loading patterns such as partial vacancy, non-agricultural products being carried, foreign objects being laid at the bottom, or mixed loading of upper and lower layers of materials can be represented in the real-time loading density distribution map as differences in local filling density values ​​and differences in regional density change values ​​in different areas, providing a readable data structure for subsequent conversion into loading density feature ratio vectors.

[0032] Specifically, in this application, when converting the real-time loading density distribution map into a loading density feature ratio vector, the map nodes in the real-time loading density distribution map can be read according to the arrangement order of the spatial grid cells in the horizontal, vertical, and longitudinal axes. The local filling density value and regional density change value corresponding to each map node are then bound to the spatial three-dimensional coordinates of that map node. This binding process ensures that the local filling density value, regional density change value, and spatial position corresponding to the same spatial grid cell remain consistent during subsequent three-dimensional grid comparisons. The loading density feature ratio vector can include multiple vector elements, each corresponding to a spatial grid cell. Each vector element can record the spatial three-dimensional coordinates, local filling density value, regional density change value, and a density feature ratio field formed by the local filling density value and regional density change value. The density feature ratio field is used to express the proportional relationship between the current spatial grid cell's filling state and the neighboring change state, and can also express the relative density state of the current spatial grid cell within the overall loading space of the carriage cavity. By converting the real-time loading density distribution map into a loading density feature ratio vector, the density distribution at different locations inside the carriage is no longer just a spatial arrangement result between map nodes, but forms a feature expression that can be compared grid-by-grid with standard density distribution feature data and can be used for loading status pattern classification.

[0033] Specifically, in this application, before performing a three-dimensional mesh comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, the method further includes: obtaining the agricultural product category identifier associated with the carriage to be inspected; retrieving pre-determined benchmark stacking density distribution data that matches the agricultural product category identifier from a multi-source configuration database; and mapping the benchmark stacking density distribution data to the same mesh dimension as the voxel data of the loading space inside the carriage to generate the standard density distribution feature data.

[0034] Specifically, in this application, the agricultural product category identifier is used to point to the agricultural product category associated with the inspection compartment. When obtaining the agricultural product category identifier associated with the inspection compartment, the category field corresponding to the inspection compartment can be read from the loading inspection task data, or the corresponding category field can be retrieved based on the inspection object identifier bound in the loading inspection trigger instruction. In this application, the agricultural product category identifier is not used to evaluate passage eligibility, but rather to determine the standard density distribution characteristic data required for subsequent 3D mesh comparison and loading status pattern classification. Different agricultural products have different dimensions, stacking gaps, surface reflection morphology, and natural stacking angles after loading, resulting in inconsistent density distributions within the compartment. For example, boxed agricultural products typically exhibit relatively regular local filling density value changes within spatial mesh cells, while bulk root and tuber agricultural products typically exhibit a layer-by-layer local filling density value distribution in the vertical direction. By selecting the corresponding baseline stacking density distribution data through the agricultural product category identifier, the loading density feature ratio vector can be compared with the spatial density baseline of the same agricultural product category, reducing the impact of natural stacking differences between different agricultural products on the determination of internal loading anomalies.

[0035] Specifically, in this application, a multi-source configuration database is used to store pre-measured data of different agricultural product categories under different carriage specifications and loading methods. The multi-source configuration database can be pre-configured through an offline calibration process. This process involves multiple carriage scans of standard loading samples of known agricultural product categories, and converting the resulting carriage interior loading space voxel data into baseline stacking density distribution data using the same statistical caliber. The baseline stacking density distribution data records the standard density values, standard density variation trends, and allowable fluctuation ranges of agricultural products in their natural stacking state at different spatial locations. The standard density value represents the filling degree of the corresponding spatial location under standard loading conditions, the standard density variation trend represents the filling variation between adjacent spatial locations under standard loading conditions, and the allowable fluctuation range absorbs differences in agricultural product shape and point cloud sampling errors. When retrieving pre-measured baseline stacking density distribution data matching the agricultural product category identifier, the agricultural product category identifier can be used as the primary search condition, combined with carriage specifications, carriage length range, carriage width range, and carriage height range to filter and obtain baseline stacking density distribution data that is closer to the carriage to be inspected. The retrieved baseline stack density distribution data is then used to generate standard density distribution feature data.

[0036] Specifically, in this application, when mapping the baseline stacking density distribution data to the same grid dimension as the cargo compartment loading space voxel data, the number of horizontal, vertical, and longitudinal grids, as well as the preset grid step size, corresponding to the cargo compartment loading space voxel data can be read first. Then, the spatial positions in the baseline stacking density distribution data are scaled according to the horizontal, vertical, and vertical boundaries of the cargo compartment's internal loading volume space. After the scale transformation, each baseline spatial position in the baseline stacking density distribution data can be mapped to the spatial grid cell used in the real-time loading density distribution map, and standard density distribution feature data is formed based on the mapped spatial grid cell. Each standard grid in the standard density distribution feature data records the same horizontal, vertical, and longitudinal coordinate intervals as the corresponding spatial grid cell in the loading density feature ratio vector, and records the standard density value and standard density change field for the corresponding agricultural product category. In this way, the loaded density feature ratio vector and the standard density distribution feature data have the same grid dimension when compared with the subsequent three-dimensional grid, and the spatial density difference can be formed in the same three-dimensional spatial coordinates, without the need to perform coordinate conversion when judging anomalies.

[0037] Specifically, in this application, when performing a three-dimensional mesh comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, the spatial three-dimensional coordinates bound to each vector element in the loading density feature ratio vector can be correlated with the standard grid at the same position in the standard density distribution feature data. The spatial three-dimensional coordinates can be expressed jointly by the center of the horizontal coordinate interval, the center of the vertical coordinate interval, and the center of the vertical coordinate interval of the spatial grid cell, or jointly by the horizontal coordinate interval, the vertical coordinate interval, and the vertical coordinate interval of the spatial grid cell. The three-dimensional mesh comparison does not only compare the overall density sum, but also compares the local filling density value, the regional density change value, the density feature ratio field, and the standard density value or standard density change field at the corresponding position in each spatial grid cell, thereby forming a spatial density difference. The spatial density difference can be recorded in the spatial three-dimensional coordinates of the spatial grid cell, the local filling density value in the loading density feature ratio vector, the regional density change value in the loading density feature ratio vector, the density feature ratio field in the loading density feature ratio vector, and the standard density value in the standard density distribution feature data. Once spatial density differences are formed, anomaly determination results can be generated based on the offset state, empty state, or abrupt change state in the spatial density differences, so that the judgment of whether there are internal loading anomalies in the carriage under inspection is directly related to the spatial location and feature ratio expression inside the carriage.

[0038] Specifically, in this application, the step of classifying the loading status of the carriage under inspection based on the spatial density difference generated by comparison, and determining whether the carriage under inspection has internal loading anomalies to generate an anomaly determination result, includes: extracting the spatial three-dimensional coordinates, regional density change values, and standard density values ​​of the standard density distribution characteristic data of the corresponding agricultural product category under the same spatial three-dimensional coordinates from the spatial density difference; calculating the actual loading centroid coordinates of the carriage under inspection based on the spatial three-dimensional coordinates and regional density change values ​​of each spatial grid unit; calculating the theoretical loading centroid coordinates of the carriage under inspection based on the spatial three-dimensional coordinates and standard density values ​​of each spatial grid unit; comparing the actual loading centroid coordinates with the theoretical loading centroid coordinates to determine the centroid offset vector; if the offset length corresponding to the centroid offset vector exceeds a set offset determination threshold, the loading status of the carriage under inspection is classified as a centroid offset anomaly mode, the carriage under inspection is determined to have internal loading anomalies, and an anomaly determination result indicating the existence of internal loading anomalies is generated.

[0039] Specifically, in this application, when extracting spatial three-dimensional coordinates, regional density change values, and standard density values ​​from the spatial density differences, the spatial three-dimensional coordinates are used to locate the position of the spatial grid unit within the loading volume space inside the carriage to be inspected. The regional density change value is used to express the strength of loading changes between the current spatial grid unit and adjacent spatial grid units. The standard density value is used to express the standard filling state of the corresponding agricultural product category under the same spatial three-dimensional coordinates. Since the regional density change value originates from the difference in changes between local filling density values, when the difference in changes has directionality, the regional density change value can be first converted into a non-negative change amplitude, and then the non-negative change amplitude can be used as a weighted component for calculating the actual loading centroid coordinates. In this way, the regional density change value can reflect the location where the loading state changes are concentrated near the spatial grid unit, and can also avoid the mutual cancellation of change differences in opposite directions in the coordinate calculation. The standard density value itself is a non-negative density field and can be directly used as a weighted component for calculating the theoretical loading centroid coordinates. When there is an allowable fluctuation range in the standard density distribution characteristic data, the standard density value can be smoothed within the allowable fluctuation range first, and then the processed standard density value can be used in the calculation of the theoretical loading centroid coordinates. Through the above processing, the actual loading centroid coordinates and the theoretical loading centroid coordinates use the same three-dimensional spatial coordinate system, and there will be no problem of inconsistency in coordinate dimensions when comparing the two.

[0040] Specifically, in this application, when calculating the actual loading centroid coordinates of the carriage to be inspected based on the spatial three-dimensional coordinates corresponding to each spatial grid unit and the regional density change value, a binding relationship can be established between the spatial three-dimensional coordinates of each spatial grid unit and the regional density change value corresponding to that spatial grid unit. Then, the spatial three-dimensional coordinates are weighted and synthesized according to the proportion of the regional density change value in all spatial grid units. The actual loading centroid coordinates may include lateral coordinate components, longitudinal coordinate components, and vertical coordinate components. The lateral coordinate component represents the position of the current concentrated loading change area of ​​the carriage to be inspected in the carriage width direction, the longitudinal coordinate component represents the position of the current concentrated loading change area of ​​the carriage to be inspected in the carriage length direction, and the vertical coordinate component represents the position of the current concentrated loading change area of ​​the carriage to be inspected in the carriage height direction. When calculating the theoretical loading centroid coordinates of the carriage to be inspected based on the spatial three-dimensional coordinates corresponding to each spatial grid unit and the standard density value, a binding relationship can be established between the spatial three-dimensional coordinates of each spatial grid unit and the standard density value under the same spatial three-dimensional coordinates. Then, the spatial three-dimensional coordinates are weighted and synthesized according to the proportion of the standard density value in all spatial grid units. The theoretical loaded centroid coordinates also include lateral, longitudinal, and vertical coordinate components. These coordinates are used to express the density center position of the corresponding agricultural product category under standard loading conditions. Since both the actual and theoretical loaded centroid coordinates are calculated using the same set of three-dimensional spatial coordinates, the subsequent centroid offset vector can be directly formed from the difference in their coordinate components.

[0041] Specifically, in this application, when comparing the actual loading centroid coordinates with the theoretical loading centroid coordinates, the differences between the two in the lateral, longitudinal, and vertical coordinate components can be compared separately, and the differences in the lateral, longitudinal, and vertical coordinate components can be combined into a centroid offset vector. The lateral component of the centroid offset vector represents the offset of the current loading change center relative to the standard density center in the width direction of the carriage; the longitudinal component represents the offset of the current loading change center relative to the standard density center in the length direction of the carriage; and the vertical component represents the offset of the current loading change center relative to the standard density center in the height direction of the carriage. The offset length can be determined by the lateral, longitudinal, and vertical components of the centroid offset vector, and is used to express the overall spatial distance between the actual loading centroid coordinates and the theoretical loading centroid coordinates. The set offset judgment threshold can be pre-configured based on the carriage specifications, preset grid step size, point cloud sampling error, and the natural stacking fluctuations of the corresponding agricultural product category, and is stored together with the agricultural product category identifier and carriage specifications. If the offset length corresponding to the centroid offset vector exceeds the set offset judgment threshold, it indicates that the loading change center represented by the loading density feature ratio vector has deviated from the standard density distribution center of the corresponding agricultural product category. The interior of the carriage to be inspected has spatial manifestations of partial emptiness, partial inclusion, or loading layering. Therefore, the loading status of the carriage to be inspected is classified as a centroid offset anomaly mode, the carriage to be inspected is determined to have the internal loading anomaly, and the anomaly judgment result used to indicate the existence of the internal loading anomaly is generated.

[0042] Specifically, in this application, the anomaly determination result can record at least the following fields: the object to be inspected identifier, the anomaly determination time, the spatial density difference that triggered the anomaly determination, the centroid offset vector, the offset length, the mode classification result of the loading state, and the result field indicating whether there is an internal loading anomaly. The object to be inspected identifier can be formed by combining the inspection station identifier carried in the loading inspection trigger command with the trigger time. The anomaly determination time can be formed by the time field when the spatial density difference completes the 3D mesh comparison. The mode classification result can be formed based on the comparison result between the offset length corresponding to the centroid offset vector and the set offset determination threshold. The object to be inspected identifier in the anomaly determination result is used to associate the anomaly determination result back to the car to be inspected. The anomaly determination time is used to associate the anomaly determination result back to the loading inspection trigger command. The spatial density difference is used to record the 3D mesh comparison data that led to the formation of the anomaly determination result. The centroid offset vector is used to express the spatial offset direction of the actual loading centroid coordinates relative to the theoretical loading centroid coordinates. The offset length is used to compare with the set offset determination threshold. The mode classification result is used to express whether the loading state of the car to be inspected is classified as an abnormal mode or a normal mode. The result field is used to express whether there is an internal loading anomaly in the car to be inspected. If the offset length does not exceed the set offset judgment threshold, an anomaly judgment result can be generated to indicate that no internal loading anomaly was found; if the offset length exceeds the set offset judgment threshold, an anomaly judgment result can be generated to indicate that an internal loading anomaly exists and the loading status is classified as a centroid offset anomaly mode. After the anomaly judgment result is generated, it can be used as input for subsequent generation of visual text verification instructions, so that the spatial density difference and mode classification results generated by the 3D mesh comparison can be converted into readable carriage orientation prompts.

[0043] The second output is as follows: Specifically, in this application, based on the aforementioned spatial density differences, real-time loading density distribution maps, loading density feature ratio vectors, and standard density distribution feature data, the method can be further implemented in terms of identifying isolated cavity areas within the interior, correcting ambient light occlusion parameter gains, identifying abnormal pattern categories, generating text verification instructions, and processing 3D mesh comparisons. The above processing still revolves around the point cloud sequence of the carriage to be inspected, the voxel data of the loading space within the carriage interior, spatial mesh cells, local filling density values, regional density change values, real-time loading density distribution maps, loading density feature ratio vectors, standard density distribution feature data, and spatial density differences, ensuring that each subsequent judgment result can be traced back to the statistical basis of the number of reflection point clouds in the spatial mesh cells, and that each subsequent judgment result continues to serve as the basis for forming abnormal judgment results or text verification instructions.

[0044] Specifically, in this application, the step of classifying the loading status of the carriage under inspection based on the spatial density difference generated by comparison, and determining whether the carriage under inspection has internal loading anomalies to generate an anomaly determination result, further includes: identifying spatial grid cells in the real-time loading density distribution map with zero regional density change value, continuous distribution between adjacent spatial grid cells, and a volume greater than a preset volume threshold, and determining the spatial grid cells with zero regional density change value, continuous distribution between adjacent spatial grid cells, and a volume greater than the preset volume threshold as internal cavity isolated void regions; if the internal cavity isolated void region exists in the real-time loading density distribution map, and the theoretical density value at the corresponding position in the standard density distribution feature data is greater than zero, then the loading status of the carriage under inspection is classified as an internal cavity void anomaly mode, the carriage under inspection is determined to have internal loading anomalies, and the anomaly type of the anomaly determination result is marked as an internal cavity void anomaly mode.

[0045] Specifically, in this application, a spatial grid cell with a regional density change value of zero refers to a spatial grid cell that does not exhibit a readable density change amplitude with its adjacent spatial grid cells, or a spatial grid cell itself lacks sufficient reflective point cloud data to characterize the continuous surface of the loading object. To avoid mistaking occasional point cloud gaps within a single spatial grid cell for isolated void regions within the cavity, the regional density change value of each spatial grid cell can be read from the real-time loading density distribution map. Then, spatial grid cells with a regional density change value of zero can be connected and merged according to their adjacency relationships along the three axes of lateral, longitudinal, and vertical. During the connection and merging process, two spatial grid cells are considered to be continuously distributed between adjacent spatial grid cells only if they share a lateral boundary, longitudinal boundary, or vertical boundary; two spatial grid cells that only touch at the corners and do not share coordinate boundaries are not included in the same continuous region. After the connection and merging process, one or more candidate continuous grid regions can be obtained. Each spatial grid cell in the candidate continuous grid region comes from the real-time loading density distribution map and retains its three-dimensional spatial coordinates, local filling density value, regional density change value, and the position of its vector element in the loading density feature ratio vector. Candidate continuous grid regions are then used for volume calculation, so that the identification of isolated cavity regions is not based on a single discrete point cloud, but on the actual extent to which continuous spatial grid cells occupy the loading volume space inside the carriage to be inspected.

[0046] Specifically, in this application, a preset volume threshold is used to limit the spatial scale of isolated void areas within the cavity. The preset volume threshold can be pre-configured based on a preset grid step size, the voxel size of the voxel data of the space loaded in the carriage cavity, the shape scale of the agricultural product corresponding to the agricultural product category identifier, and the point cloud sampling error of the carriage scanning device. When configuring the preset volume threshold, the volume of a single spatial grid unit can be determined first, and then a lower limit of the volume can be set based on the allowable gap volume range of the same agricultural product category in a natural stacking state, so that the natural gaps between individual agricultural products will not directly trigger the determination of isolated void areas within the cavity. For each candidate continuous grid region, the number of spatial grid units in it can be converted to the volume of a single spatial grid unit. If the candidate continuous grid region is located at the edge of the space voxel data loaded in the carriage cavity and contains spatial grid units in the form of incomplete cuboids, the participating volume of the spatial grid unit is determined based on the actual number of voxels covered by the spatial grid unit in the space voxel data loaded in the carriage cavity, and the participating volume of the spatial grid unit is included in the volume of the candidate continuous grid region. When the volume of a candidate continuous grid region exceeds a preset volume threshold, the candidate continuous grid region is identified as an isolated cavity region within the cavity. After the isolated cavity region is identified, its spatial three-dimensional coordinate range, continuous distribution direction, volume data, and the corresponding vector element positions are retained in the anomaly detection result so that it can be converted into a carriage orientation description when generating subsequent text verification instructions.

[0047] Specifically, in this application, the theoretical density value at the corresponding position in the standard density distribution feature data is used to represent the filling state that the corresponding agricultural product category should have under the same three-dimensional spatial coordinates. The theoretical density value can be formed by reading the standard density value at the corresponding position in the standard density distribution feature data during the identification of isolated cavity areas. When mapping the isolated cavity areas to the standard density distribution feature data, the three-dimensional spatial coordinates of multiple spatial grid units covered by the isolated cavity areas can be read, and then the theoretical density value under the same three-dimensional spatial coordinates can be extracted from the standard density distribution feature data. If the theoretical density value at the corresponding position in the standard density distribution feature data is greater than zero, it indicates that under the standard loading state of the corresponding agricultural product category, this position should not be empty or a continuous, unchanging area; while the real-time loading density distribution map has already formed an isolated cavity area, it indicates that the position corresponding to the same three-dimensional spatial coordinates in the loading volume space inside the carriage has a continuous spatial missingness that is inconsistent with the standard stacking state. Based on the correspondence between this continuous spatial gap and the theoretical density value, the loading status of the inspected carriage can be classified as an internal cavity void anomaly pattern. This determines that the inspected carriage has an internal loading anomaly, and the anomaly type of the anomaly determination result is marked as an internal cavity void anomaly pattern. At the data level, the internal cavity void anomaly pattern corresponds to a continuous spatial difference between the standard density distribution characteristic data corresponding to the agricultural product category identifier associated with the inspected carriage and the real-time loading density distribution map under the same three-dimensional coordinates, rather than being inferred solely from overall weight or external dimensions.

[0048] Specifically, in this application, before comparing the loading density feature ratio vector with the standard density distribution feature data of the corresponding agricultural product category using a three-dimensional grid, the method further includes: obtaining ambient light occlusion parameters of the external environment; if the ambient light occlusion parameters meet a set disturbance range, then performing gain correction on the number of reflected point clouds in each of the spatial grid cells to compensate for the point cloud missing error caused by external environmental occlusion, and updating the real-time loading density distribution map based on the gain-corrected number of reflected point clouds; and converting the updated real-time loading density distribution map back into the loading density feature ratio vector.

[0049] Specifically, in this application, the ambient light occlusion parameter is used to describe the degree of influence of the external environment on the quality of the point cloud sequence acquisition of the carriage when it is in the loading and inspection station. The ambient light occlusion parameter can be jointly determined by the brightness distribution in the loading and inspection station image, the reflection intensity distribution in the carriage point cloud sequence, and the proportion of missing points in the same spatial grid cell in consecutive point cloud frames. The loading and inspection station image can be acquired by the image acquisition device configured at the loading and inspection station where the carriage scanning device is located during the acquisition period corresponding to the loading and inspection trigger command. The brightness distribution in the loading and inspection station image is used to reflect the local brightness changes caused by strong light, shadows, or occlusion. The reflection intensity distribution in the carriage point cloud sequence is used to reflect the stability of the point cloud reflection received by the carriage scanning device. The proportion of missing points in the same spatial grid cell in consecutive point cloud frames is used to reflect whether the number of reflected point clouds at the same spatial location has abnormally decreased during continuous sampling. The brightness distribution in the loading inspection station image, the reflection intensity distribution in the carriage point cloud sequence, and the proportion of missing points in consecutive point cloud frames for the same spatial grid cell can be generated during the acquisition period corresponding to the loading inspection trigger command and bound to the carriage point cloud sequence of the carriage to be inspected. After the ambient light occlusion parameters are formed, their values ​​do not directly participate in the internal loading anomaly determination, but are first used to determine whether it is necessary to correct the number of reflection point clouds in each spatial grid cell, so as to reduce the impact of point cloud missing errors caused by external environmental occlusion on the real-time loading density distribution map and loading density feature ratio vector.

[0050] Specifically, in this application, the set disturbance interval can be pre-configured after the loading inspection station is deployed through multiple empty carriage scans and standard loading scans. When configuring the set disturbance interval, the value range of the ambient light occlusion parameter under normal lighting conditions can be recorded, and then the value range of the ambient light occlusion parameter under conditions such as strong light, shadow, and partial occlusion can be recorded. The value range that can distinguish between the stable state of point cloud sampling and the state of missing point cloud is used as the set disturbance interval. If the ambient light occlusion parameter does not meet the set disturbance interval, the real-time loading density distribution map is generated using the statistical results of the number of reflected point clouds that have already been formed, and the loading density feature ratio vector is generated based on the real-time loading density distribution map. If the ambient light occlusion parameter meets the set disturbance interval, it means that some spatial grid cells in the carriage point cloud sequence are affected by external environmental occlusion. Directly calculating the local filling density value based on the number of reflected point clouds will cause the spatial density difference to carry point cloud missing error. At this point, the number of reflected point clouds within each spatial grid cell can be adjusted by gain correction. The gain coefficient for gain correction can be determined based on the attenuation of reflection intensity at the location of the spatial grid cell, the trend of the number of reflected point clouds in adjacent spatial grid cells, and the proportion of missing point clouds in consecutive point cloud frames. The gain coefficient is used to adjust the number of reflected point clouds affected by external environmental occlusion to a data range that matches the trend of the number of reflected point clouds in adjacent spatial grid cells. The adjusted number of reflected point clouds is then re-involved in the calculation of local fill density values. The updated local fill density values ​​are used to calculate regional density change values, and the real-time loading density distribution map is updated based on the updated regional density change values. After the real-time loading density distribution map is updated, the updated local fill density values ​​and updated regional density change values ​​are re-positioned according to the arrangement order of the spatial grid cells in the horizontal, vertical, and longitudinal axes, thereby converting the updated real-time loading density distribution map back into a loading density feature ratio vector. In this way, the loading density feature ratio vector has already compensated for the point cloud missing error indicated by the ambient light occlusion parameters before entering the 3D grid comparison.

[0051] Specifically, in this application, when the anomaly determination result indicates the existence of an internal loading anomaly, the method further includes: identifying the anomaly pattern category corresponding to the internal loading anomaly based on the grid distribution area that generates the spatial density difference between the real-time loading density distribution map and the standard density distribution feature data; wherein, the anomaly pattern category includes at least one of the following: empty front of the carriage, non-agricultural products sandwiched in the middle of the carriage, foreign objects laid at the bottom of the carriage, and mixed loading of upper and lower layers of materials.

[0052] Specifically, in this application, the grid distribution area that generates the spatial density difference originates from the three-dimensional grid comparison result between the loading density feature ratio vector and the standard density distribution feature data. This grid distribution area can be composed of spatial grid units whose spatial density differences exceed the allowable fluctuation range in the standard density distribution feature data, and retains the arrangement positions of these spatial grid units in the horizontal, vertical, and lateral axes. When identifying the abnormal pattern category, the grid distribution area can first be divided into the front area, middle area, and rear area of ​​the carriage according to its vertical position, and then into the bottom area, middle layer area, and upper layer area of ​​the carriage according to its vertical position. Subsequently, the abnormal pattern category is determined based on the continuity of the grid distribution area in different carriage orientations, the direction of density difference, and the density feature ratio field in the corresponding vector element. The abnormal pattern category here is not directly generated from textual registration information, but is identified by the distribution position and distribution pattern of the spatial density difference in the loading volume space inside the carriage to be inspected. Therefore, it can form a correspondence with the map nodes in the real-time loading density distribution map and the vector elements in the loading density feature ratio vector.

[0053] Specifically, in this application, if the grid distribution area causing the spatial density difference is mainly concentrated in the front area of ​​the carriage, and the local filling density value of the real-time loading density distribution map in the front area of ​​the carriage is lower than the standard density value in the same three-dimensional coordinates in the standard density distribution feature data, and the front area of ​​the carriage is continuously distributed in the horizontal or vertical direction, then the abnormal pattern category corresponding to the internal loading anomaly can be identified as the front of the carriage being empty. If the grid distribution area causing the spatial density difference is located in the middle area of ​​the carriage, and the difference between the local filling density value and the standard density value in the middle area of ​​the carriage presents a local blocky concentration, and the density change value at the boundary of the local blocky area is higher than that of the surrounding spatial grid cells, then the abnormal pattern category corresponding to the internal loading anomaly can be identified as the middle of the carriage carrying non-agricultural products pattern. If the grid distribution area causing the spatial density difference is located in the bottom area of ​​the carriage, and there is an abnormal density difference that extends continuously along the length of the carriage in the bottom area of ​​the carriage, and the local filling density value of the adjacent spatial grid cells above is still close to the corresponding standard density value, then the abnormal pattern category corresponding to the internal loading anomaly can be identified as the bottom of the carriage being covered with foreign objects pattern. If the grid distribution areas that generate the spatial density differences are vertically distributed between the bottom, middle, and upper layers of the carriage, and the density variation values ​​between different vertical layers exhibit abrupt changes between layers, then the abnormal pattern category corresponding to the internal loading anomaly can be identified as a layered mixed loading pattern of upper and lower layers of materials. The identified abnormal pattern category is then written into the anomaly determination result, so that the anomaly determination result not only indicates whether an internal loading anomaly exists, but also expresses the pattern type of the internal loading anomaly in the loading volume space inside the carriage to be inspected.

[0054] Specifically, in this application, the step of generating a visualized text verification instruction based on the anomaly determination result and pushing the text verification instruction to an external inspection terminal for manual anomaly verification includes: extracting the three-dimensional coordinate range of the spatial grid cell where the internal loading anomaly is located based on the current anomaly determination result; converting the three-dimensional coordinate range of the spatial grid cell where the internal loading anomaly is located into a text field indicating the specific location of the carriage for verification, and embedding the text field into a preset text template to generate the text verification instruction; establishing a wireless communication link with the external inspection terminal based on the network node identifier of the current loading inspection station, and sending the text verification instruction to the screen of the external inspection terminal for highlighting.

[0055] Specifically, in this application, the three-dimensional coordinate interval of the spatial grid cell where the internal loading anomaly is located is derived from the spatial density difference, isolated cavity area, centroid offset vector, anomaly mode category, and corresponding loading density feature ratio vector elements recorded in the anomaly judgment result. When extracting the three-dimensional coordinate interval, the spatial grid cells marked as anomalies in the anomaly judgment result can be read first, and then the minimum coverage area can be calculated based on the horizontal, vertical, and lateral coordinate intervals of these spatial grid cells. This minimum coverage area corresponds to the left, middle, or right side of the carriage horizontally, the front, middle, or rear of the carriage vertically, and the bottom, middle, or upper layer of the carriage vertically. Through this coordinate interval extraction method, the three-dimensional data in the anomaly judgment result can be converted into carriage location information that can be directly located during on-site verification, without requiring the external inspection terminal to re-parse all map nodes of the real-time loading density distribution map or all vector elements of the loading density feature ratio vector.

[0056] Specifically, in this application, when converting the three-dimensional coordinate range of the spatial grid cell where the internal loading anomaly is located into a text field indicating the specific location for verification within the carriage, the location description can be generated according to the sequential relationship of the horizontal coordinate range, the vertical coordinate range, and the lateral coordinate range. For example, if the three-dimensional coordinate range falls in the longitudinal front and vertical bottom of the carriage, the text field can be expressed as "verifying the bottom area of ​​the front of the carriage"; if the three-dimensional coordinate range falls in the longitudinal middle and lateral left of the carriage, the text field can be expressed as "verifying the left side area of ​​the middle of the carriage". The preset text template can be pre-configured to include fields such as the object to be inspected, the anomaly pattern category, the carriage location description, the verification focus, and the prompt level. The verification focus can be formed by the anomaly pattern category in the anomaly judgment result, and the prompt level can be formed by the deviation of the centroid offset vector, the volume data of the isolated cavity area, or the density feature ratio field. When text fields are embedded into a preset text template, the text fields are used to fill the carriage orientation description field, the anomaly pattern category in the anomaly judgment result is used to fill the review focus field, and the deviation of the centroid offset vector, the volume data of isolated cavities in the cavity, or the density feature ratio field can be used to fill the prompt level field. The resulting text review instructions can transform spatial density differences and pattern classification results into text content readable by external inspection terminals, establishing a transferable data relationship between anomaly judgment results and on-site review actions.

[0057] Specifically, in this application, the network node identifier of the current loading and inspection station can be recorded when the loading and inspection trigger command is generated, or it can be read by the loading and inspection station where the carriage to be inspected is located. The network node identifier is used to indicate the communication entry corresponding to the current loading and inspection station and is not used to describe the road traffic control status. When establishing a wireless communication link with the external inspection terminal, the communication address corresponding to the current loading and inspection station can be selected according to the network node identifier, and then a message payload containing a text review command can be sent to the external inspection terminal. The message payload may include the identifier of the object to be inspected, the text review command, the anomaly judgment result, and the highlighted display field. After receiving the message payload, the external inspection terminal can display the text review command in the prompt area of ​​the screen according to the highlighted display field, so that the external inspection terminal can present the carriage location description and anomaly mode category corresponding to the carriage to be inspected. When the wireless communication link fails to be established, the text review command can be written into the text review command temporary storage record, and the text review command in the text review command temporary storage record can be resent according to the network node identifier to avoid the anomaly judgment result remaining in the background data state without entering the on-site review and display stage.

[0058] Specifically, in this application, the step of performing a three-dimensional mesh comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category includes: reading the spatial three-dimensional coordinates, local filling density values, and regional density change values ​​bound to multiple spatial grid units in the loading density feature ratio vector; reading the standard density distribution feature data of the corresponding agricultural product category, and extracting the standard density field and standard density change field corresponding to each spatial grid unit from the standard density distribution feature data based on the spatial three-dimensional coordinates; aligning the spatial three-dimensional coordinates of each spatial grid unit with the grid coordinates in the standard density distribution feature data to establish the loading density feature ratio vector. The grid correspondence between the quantity and the standard density distribution feature data is established; based on the grid correspondence, the local filling density value under the same spatial three-dimensional coordinates is differentially processed with the standard density field, and the regional density change value under the same spatial three-dimensional coordinates is differentially processed with the standard density change field to form unit density difference data corresponding to each spatial grid unit; according to the arrangement order of the spatial grid units in the horizontal, vertical and vertical axes, the multiple unit density difference data are positionally bound to generate the spatial density difference used to characterize the difference distribution state between the loading density feature ratio vector and the standard density distribution feature data.

[0059] Specifically, in this application, when reading the spatial three-dimensional coordinates, local filling density values, and regional density change values ​​bound to multiple spatial grid cells in the loading density feature ratio vector, the spatial grid cells can be read sequentially according to their index order in the horizontal, vertical, and axial directions. The spatial three-dimensional coordinates corresponding to each spatial grid cell can be represented by the horizontal, vertical, and axial coordinate intervals of that spatial grid cell. The local filling density value comes from the statistical results of the number of reflected point clouds within that spatial grid cell, and the regional density change value comes from the difference in local filling density values ​​between that spatial grid cell and its adjacent spatial grid cells. The obtained spatial three-dimensional coordinates, local filling density values, and regional density change values ​​together constitute the vector reading data of the loading density feature ratio vector. The vector reading data retains information on three aspects: spatial location, local filling degree, and neighborhood change degree. When extracting the standard density field and standard density change field from the standard density distribution feature data subsequently, the same spatial three-dimensional coordinates can be used as the retrieval basis.

[0060] Specifically, in this application, the standard density field represents the standard filling density of the corresponding agricultural product category under the same three-dimensional spatial coordinates, and the standard density variation field represents the degree of standard variation between the corresponding agricultural product category and adjacent standard grids under the same three-dimensional spatial coordinates. When reading the standard density distribution feature data of the corresponding agricultural product category, the standard density distribution feature data mapped to the same grid dimension of the loading space voxel data in the carriage cavity can be determined first based on the agricultural product category identifier. Then, the corresponding grids in the standard density distribution feature data are retrieved one by one according to the three-dimensional spatial coordinates in the loading density feature vector. If the grid coordinates in the standard density distribution feature data are completely consistent with the three-dimensional spatial coordinates, the standard density field and the standard density variation field are directly extracted. If there are non-complete cuboid-shaped spatial grid units at the edge position of the loading space voxel data in the carriage cavity where the grid coordinates in the standard density distribution feature data and the three-dimensional spatial coordinates are located, the correspondence is determined by the overlapping range of the coordinate intervals of the spatial grid units, and the standard density field and the standard density variation field are extracted within the overlapping range. The extracted standard density field and the standard density variation field, together with the vector reading data, enter the grid correspondence establishment process.

[0061] Specifically, in this application, when aligning the spatial three-dimensional coordinates of each spatial grid cell with the grid coordinates in the standard density distribution feature data, the same horizontal coordinate interval, the same vertical coordinate interval, and the same vertical coordinate interval can be used as alignment conditions. After alignment, a vector element in the loading density feature ratio vector forms a one-to-one grid correspondence with a standard grid in the standard density distribution feature data. The grid correspondence records at least the spatial three-dimensional coordinates, local filling density value, regional density change value, standard density field, and standard density change field, and may also record the density feature ratio field in the corresponding vector element. Since the standard density distribution feature data has been mapped to the same grid dimension as the loading space voxel data of the carriage cavity, the grid correspondence can be directly established in the same spatial three-dimensional coordinates, avoiding the problem of inconsistent dimensions and spatial positions caused by directly subtracting grid data of different sizes. After the grid correspondence is formed, subsequent unit density difference data can be calculated independently within each spatial grid cell.

[0062] Specifically, in this application, when forming unit density difference data based on the grid correspondence, the local filling density value and standard density field under the same three-dimensional coordinates can be read first, and the difference between the local filling density value and the standard density field can be processed to obtain the filling density difference of the spatial grid unit. Subsequently, the regional density change value and standard density change field under the same three-dimensional coordinates can be read, and the difference between the regional density change value and the standard density change field can be processed to obtain the change density difference of the spatial grid unit. The filling density difference is used to express the difference between the actual filling state of the current spatial grid unit and the standard filling state of the corresponding agricultural product category, and the change density difference is used to express the difference between the actual degree of change and the standard degree of change between the current spatial grid unit and its neighboring spaces. After binding the filling density difference, change density difference, and spatial three-dimensional coordinates, the unit density difference data corresponding to the spatial grid unit can be formed. The unit density difference data is retained at the spatial grid unit level, so that the subsequent anomaly judgment results can locate the specific spatial grid unit, rather than only giving the overall difference value.

[0063] Specifically, in this application, when performing position binding processing on multiple unit density difference data, each unit density difference data can be written to its corresponding three-dimensional position according to the arrangement order of the spatial grid units in the three axes of horizontal, vertical, and longitudinal. The horizontal arrangement order is used to maintain the left-right positional relationship in the width direction of the carriage, the longitudinal arrangement order is used to maintain the front-back positional relationship in the length direction of the carriage, and the longitudinal arrangement order is used to maintain the upper-lower layer positional relationship in the height direction of the carriage. After the position binding processing is completed, multiple unit density difference data together form a spatial density difference. The spatial density difference includes not only the filling density difference of each spatial grid unit, but also the variation density difference of each spatial grid unit, as well as the three-dimensional arrangement relationship of these differences in the loading volume space inside the carriage to be inspected. The spatial density difference thus formed can be used to identify centroid offset anomaly patterns, internal cavity cavity anomaly patterns, and other anomaly pattern categories, and can serve as the basic data for determining the carriage inspection orientation when generating text inspection instructions.

[0064] like Figure 3 As shown, this is an embodiment of the present application of a device for identifying and inspecting loading anomalies based on density feature ratio, which includes: The carriage point cloud sequence acquisition module is configured to acquire the carriage point cloud sequence of the carriage to be inspected in response to the loading inspection trigger command of the carriage to be inspected; The voxel data construction module is configured to perform coordinate axis calibration on the point cloud sequence of the carriage and remove the body shell feature data and wheel feature data from the point cloud sequence of the carriage to construct the voxel data of the loading space inside the carriage cavity of the carriage to be inspected. The density feature ratio vector generation module is configured to divide the voxel data of the loading space inside the carriage into multiple spatial grid cells, count the local filling density value in each spatial grid cell, and determine the regional density change value based on the difference in the local filling density value between adjacent spatial grid cells, so as to generate a real-time loading density distribution map of the carriage to be inspected, and convert the real-time loading density distribution map into a loading density feature ratio vector. The anomaly determination result generation module is configured to perform a three-dimensional grid comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, and to perform pattern classification on the loading status of the carriage to be inspected based on the spatial density difference generated by the comparison, thereby determining whether there is an internal loading anomaly in the carriage to be inspected, so as to generate an anomaly determination result. The text verification instruction push module is configured to generate a visual text verification instruction based on the anomaly determination result when the anomaly determination result indicates that there is an internal loading anomaly, and push the text verification instruction to an external verification terminal for manual verification of the anomaly.

[0065] like Figure 4 As shown, an electronic device includes a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the loading anomaly identification and verification method based on density feature ratio as described in any one of the claims of this application.

[0066] like Figure 5 As shown, this is a computer-readable storage medium according to an embodiment of the present application. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the loading anomaly identification and verification method based on density feature ratio as described in any one of the present applications.

[0067] like Figure 6 As shown, this is an embodiment of the present application of a loading anomaly identification and inspection system based on density feature ratio, which includes a point cloud acquisition device, an anomaly pattern classification and identification device, and an external inspection terminal. The point cloud acquisition device is used to respond to the loading inspection trigger command of the carriage to be inspected, acquire the carriage point cloud sequence of the carriage to be inspected, and send the carriage point cloud sequence to the abnormal pattern classification and recognition device. The abnormal pattern classification and recognition device is used to perform coordinate axis calibration on the carriage point cloud sequence and remove the body shell feature data and wheel feature data from the carriage point cloud sequence in order to construct the carriage interior loading space voxel data of the carriage to be inspected. The abnormal pattern classification and recognition device is also used to divide the voxel data of the loading space inside the carriage into multiple spatial grid units, count the local filling density value in each spatial grid unit, and determine the regional density change value based on the difference in the local filling density value between adjacent spatial grid units, so as to generate a real-time loading density distribution map of the carriage to be inspected, and convert the real-time loading density distribution map into a loading density feature ratio vector. The abnormal pattern classification and recognition device is also used to perform a three-dimensional grid comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, and to perform pattern classification on the loading status of the carriage to be inspected based on the spatial density difference generated by the comparison, to determine whether there is an internal loading abnormality in the carriage to be inspected, so as to generate an abnormality judgment result. The abnormal pattern classification and recognition device is also used to generate a visual text verification instruction based on the abnormal judgment result when the abnormal judgment result indicates that there is an internal loading abnormality, and push the text verification instruction to the external verification terminal. The external verification terminal is used to receive the text verification instruction and to visually display the text verification instruction.

[0068] The above Figures 3-6 For an exemplary description, please refer to the above. Figure 1 This will not be elaborated upon here.

Claims

1. A method for identifying and verifying loading anomalies based on density feature ratio, characterized in that, include: In response to the loading inspection trigger command of the carriage to be inspected, the point cloud sequence of the carriage to be inspected is obtained; The coordinate axis of the carriage point cloud sequence is calibrated, and the body shell feature data and wheel feature data in the carriage point cloud sequence are removed to construct the carriage interior loading space voxel data of the carriage to be inspected. The voxel data of the loading space inside the carriage is divided into multiple spatial grid cells. The local filling density value in each spatial grid cell is counted. The regional density change value is determined based on the difference in the local filling density value between adjacent spatial grid cells to generate a real-time loading density distribution map of the carriage to be inspected. The real-time loading density distribution map is then converted into a loading density feature ratio vector. The loading density feature vector is compared with the standard density distribution feature data of the corresponding agricultural product category in a three-dimensional grid. Based on the spatial density difference generated by the comparison, the loading status of the inspection compartment is classified into patterns to determine whether there is an internal loading abnormality in the inspection compartment, so as to generate an abnormality judgment result. When the anomaly determination result indicates the existence of an internal loading anomaly, a visual text verification instruction is generated based on the anomaly determination result, and the text verification instruction is pushed to an external verification terminal for manual verification of the anomaly.

2. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 1, characterized in that, The process of calibrating the coordinate axes of the carriage point cloud sequence and removing body shell feature data and wheel feature data from the carriage point cloud sequence to construct the carriage interior loading space voxel data of the carriage to be inspected includes: Establish the initial three-dimensional coordinate values ​​of the carriage point cloud sequence in a spatial rectangular coordinate system; The longitudinal centerline of the carriage to be inspected is determined by using the principal component directions of the initial three-dimensional coordinate values, and the initial three-dimensional coordinate values ​​are transformed into a standardized coordinate system based on the longitudinal centerline of the carriage to complete the coordinate axis calibration of the carriage point cloud sequence. Based on a preset spatial boundary contour threshold, point cloud points belonging to external physical components of the carriage are identified in the carriage point cloud sequence. The body shell feature data and wheel feature data are determined based on the point cloud points of the external physical components of the carriage. The body shell feature data and wheel feature data are removed from the carriage point cloud sequence after coordinate axis calibration. The carriage interior loading space voxel data are constructed based on the retained internal loading volume space of the carriage to be inspected.

3. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 1, characterized in that, The process involves dividing the voxel data of the loading space inside the carriage into multiple spatial grid cells, statistically analyzing the local filling density values ​​within each spatial grid cell, and determining the regional density change value based on the difference in local filling density values ​​between adjacent spatial grid cells to generate a real-time loading density distribution map of the carriage under inspection. The real-time loading density distribution map is then converted into a loading density feature ratio vector, including: According to the preset grid step size, the voxel data of the loading space in the inner cavity of the carriage is divided into voxels in the three axes of horizontal, vertical and vertical directions to form multiple cuboid-shaped spatial grid units. The number of reflected point clouds contained in each spatial grid cell is counted, and the local filling density value of the corresponding spatial grid cell is determined based on the number of reflected point clouds. The variation difference is calculated based on the local filling density value between two adjacent spatial grid cells, and the regional density variation value of the corresponding spatial grid cell is determined based on the variation difference, so as to generate the real-time loading density distribution map based on the regional density variation values ​​corresponding to multiple spatial grid cells; According to the arrangement order of the spatial grid cells in the three axes of horizontal, vertical and longitudinal, the local filling density value and regional density change value corresponding to each spatial grid cell are subjected to position binding processing to convert the real-time loading density distribution map into the loading density feature ratio vector.

4. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 1, characterized in that, Before performing a three-dimensional mesh comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, the method further includes: Obtain the agricultural product category identifier associated with the carriage to be inspected; Retrieve pre-determined baseline stacking density distribution data that matches the agricultural product category identifier from a multi-source configuration database; The baseline stacking density distribution data is mapped to the same grid dimension as the voxel data of the loading space inside the carriage to generate the standard density distribution feature data.

5. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 1, characterized in that, The step of classifying the loading status of the carriage under inspection based on the spatial density differences generated by comparison, and determining whether there are internal loading anomalies in the carriage under inspection to generate an anomaly determination result, includes: Extract the spatial three-dimensional coordinates, regional density change values, and standard density values ​​of the corresponding agricultural product categories from the spatial density differences, and extract the standard density distribution characteristic data of each spatial grid unit under the same spatial three-dimensional coordinates. Based on the spatial three-dimensional coordinates of each spatial grid unit and the regional density change value, the actual loading centroid coordinates of the carriage to be inspected are calculated. Based on the spatial three-dimensional coordinates corresponding to each of the spatial grid units and the standard density value, calculate the theoretical loading centroid coordinates of the carriage to be inspected; The actual loaded centroid coordinates are compared with the theoretical loaded centroid coordinates to determine the centroid offset vector; If the offset length corresponding to the centroid offset vector exceeds the set offset judgment threshold, the loading status of the car under inspection is classified as a centroid offset abnormal mode, the car under inspection is determined to have an internal loading abnormality, and an abnormality judgment result is generated to indicate the existence of an internal loading abnormality.

6. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 5, characterized in that, The step of classifying the loading status of the carriage under inspection based on the spatial density difference generated by comparison, determining whether there is an internal loading anomaly in the carriage under inspection, and generating an anomaly determination result, further includes: Identify spatial grid cells in the real-time loading density distribution map that have zero regional density change, are continuously distributed between adjacent spatial grid cells, and have a volume greater than a preset volume threshold, and determine the spatial grid cells that have zero regional density change, are continuously distributed between adjacent spatial grid cells, and have a volume greater than a preset volume threshold as isolated cavity regions within the cavity; If an isolated cavity region exists in the real-time loading density distribution map, and the theoretical density value at the corresponding position in the standard density distribution feature data is greater than zero, then the loading status of the carriage under inspection is classified as an internal cavity cavity anomaly mode, the carriage under inspection is determined to have an internal loading anomaly, and the anomaly type of the anomaly determination result is marked as an internal cavity cavity anomaly mode.

7. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 1, characterized in that, Before performing a three-dimensional mesh comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, the method further includes: Obtain ambient light occlusion parameters of the external environment; If the ambient light occlusion parameter meets the set disturbance range, the number of reflected point clouds in each of the spatial grid cells is adjusted by gain to compensate for the point cloud missing error caused by external environmental occlusion, and the real-time loading density distribution map is updated according to the number of reflected point clouds after gain adjustment. The updated real-time loading density distribution map is then converted back into the loading density feature ratio vector.

8. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 1, characterized in that, When the anomaly determination result indicates the presence of an internal loading anomaly, the method further includes: Based on the grid distribution region that generates the spatial density difference between the real-time loading density distribution map and the standard density distribution feature data, the abnormal pattern category corresponding to the internal loading anomaly is identified. The abnormal mode categories include at least one of the following: empty front of the carriage, non-agricultural products mixed in the middle of the carriage, foreign objects laid at the bottom of the carriage, and mixed loading of materials in the upper and lower layers.

9. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 1, characterized in that, The step of generating a visualized text verification instruction based on the anomaly determination result and pushing the text verification instruction to an external verification terminal for manual anomaly verification includes: Based on the current anomaly determination result, extract the three-dimensional coordinate range of the spatial grid cell where the internal loading anomaly is located; The three-dimensional coordinate range of the spatial grid cell where the internal loading anomaly is located is converted into a text field indicating the specific location of the carriage for verification, and the text field is embedded into a preset text template to generate the text verification instruction. Based on the network node identifier of the current loading and inspection station, a wireless communication link is established with the external inspection terminal, and the text verification instruction is sent to the screen of the external inspection terminal for highlighting.

10. The method for identifying and verifying loading anomalies based on density feature ratio according to claim 1, characterized in that, The step of performing a three-dimensional grid comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category includes: Read the spatial three-dimensional coordinates, local fill density values, and regional density variation values ​​that are bound to multiple spatial grid cells in the loading density feature ratio vector; Read the standard density distribution feature data of the corresponding agricultural product category, and extract the standard density field and standard density change field corresponding to each spatial grid unit from the standard density distribution feature data according to the spatial three-dimensional coordinates; Align the spatial three-dimensional coordinates of each of the spatial grid cells with the grid coordinates in the standard density distribution feature data to establish the grid correspondence between the loaded density feature ratio vector and the standard density distribution feature data; Based on the grid correspondence, the local filling density value under the same spatial three-dimensional coordinates is compared with the standard density field, and the regional density change value under the same spatial three-dimensional coordinates is compared with the standard density change field to form the unit density difference data corresponding to each spatial grid unit; According to the arrangement order of the spatial grid cells in the three axes of horizontal, vertical and longitudinal, the density difference data of multiple cells are subjected to position binding processing to generate the spatial density difference used to characterize the difference distribution state between the loading density feature ratio vector and the standard density distribution feature data.

11. A device for identifying and inspecting loading anomalies based on density feature ratio, characterized in that, include: The carriage point cloud sequence acquisition module is configured to acquire the carriage point cloud sequence of the carriage to be inspected in response to the loading inspection trigger command of the carriage to be inspected; The voxel data construction module is configured to perform coordinate axis calibration on the point cloud sequence of the carriage and remove the body shell feature data and wheel feature data from the point cloud sequence of the carriage to construct the voxel data of the loading space inside the carriage cavity of the carriage to be inspected. The density feature ratio vector generation module is configured to divide the voxel data of the loading space inside the carriage into multiple spatial grid cells, count the local filling density value in each spatial grid cell, and determine the regional density change value based on the difference in the local filling density value between adjacent spatial grid cells, so as to generate a real-time loading density distribution map of the carriage to be inspected, and convert the real-time loading density distribution map into a loading density feature ratio vector. The anomaly determination result generation module is configured to perform a three-dimensional grid comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, and to perform pattern classification on the loading status of the carriage to be inspected based on the spatial density difference generated by the comparison, thereby determining whether there is an internal loading anomaly in the carriage to be inspected, so as to generate an anomaly determination result. The text verification instruction push module is configured to generate a visual text verification instruction based on the anomaly determination result when the anomaly determination result indicates that there is an internal loading anomaly, and push the text verification instruction to an external verification terminal for manual verification of the anomaly.

12. An electronic device, characterized in that, The device includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the loading anomaly identification and verification method based on density feature ratio as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the loading anomaly identification and verification method based on density feature ratio as described in any one of claims 1 to 10.

14. A system for identifying and verifying loading anomalies based on density feature ratio, characterized in that, This includes point cloud acquisition equipment, abnormal pattern classification and recognition equipment, and external inspection terminals; The point cloud acquisition device is used to respond to the loading inspection trigger command of the carriage to be inspected, acquire the carriage point cloud sequence of the carriage to be inspected, and send the carriage point cloud sequence to the abnormal pattern classification and recognition device. The abnormal pattern classification and recognition device is used to perform coordinate axis calibration on the carriage point cloud sequence and remove the body shell feature data and wheel feature data from the carriage point cloud sequence in order to construct the carriage interior loading space voxel data of the carriage to be inspected. The abnormal pattern classification and recognition device is also used to divide the voxel data of the loading space inside the carriage into multiple spatial grid units, count the local filling density value in each spatial grid unit, and determine the regional density change value based on the difference in the local filling density value between adjacent spatial grid units, so as to generate a real-time loading density distribution map of the carriage to be inspected, and convert the real-time loading density distribution map into a loading density feature ratio vector. The abnormal pattern classification and recognition device is also used to perform a three-dimensional grid comparison between the loading density feature ratio vector and the standard density distribution feature data of the corresponding agricultural product category, and to perform pattern classification on the loading status of the carriage to be inspected based on the spatial density difference generated by the comparison, to determine whether there is an internal loading abnormality in the carriage to be inspected, so as to generate an abnormality judgment result. The abnormal pattern classification and recognition device is also used to generate a visual text verification instruction based on the abnormal judgment result when the abnormal judgment result indicates that there is an internal loading abnormality, and push the text verification instruction to the external verification terminal. The external verification terminal is used to receive the text verification instruction and to visually display the text verification instruction.

Images