Bulk feed truck unloading flow detection method and system based on visual recognition

Abstract

A bulk feed truck unloading flow detection method and system based on visual recognition, in the method, the reference density is calculated based on the full load initial mass and volume, and the real-time spatial form of the feed surface is obtained periodically during the unloading process. By comparing the spatial boundary changes of adjacent periods and the initial state, the incremental discharge volume and the cumulative discharge volume are calculated respectively. Then, the unloading space evolution ratio is obtained by using the ratio of the cumulative discharge volume to the initial total volume, and the density compensation coefficient is mapped to dynamically correct the reference density to obtain the instantaneous equivalent density. Finally, the incremental unloading mass of the current period is obtained by multiplying the incremental discharge volume and the instantaneous equivalent density, and the real-time cumulative unloading total mass is finally obtained by accumulation. The application is used to improve the accuracy of unloading mass flow detection, and further improve the refinement degree of the intelligent monitoring system of agricultural products in the precise feeding program control of feed.

Description

Technical Field

[0001] This application belongs to the field of intelligent monitoring of agricultural products, and in particular relates to a method and system for detecting the unloading flow of bulk feed trucks based on visual recognition. Background Technology

[0002] Accurate detection of feed truck unloading flow rate is a crucial step in achieving scientific feed formulation and programmed automatic control. Currently, the industry typically employs a volume measurement technology based on three-dimensional topographic scanning of the truck's interior. This technology involves installing a three-dimensional time-of-flight camera fixed to the top inner wall of the enclosed bulk feed truck compartment. During the unloading process by the auger at the bottom, the camera continuously performs high-frequency scanning of the top surface of the feed inside the compartment, acquiring real-time three-dimensional point cloud data of the feed surface and reconstructing its shape. The system accurately calculates the real-time reduction rate of feed volume based on the decrease in feed surface height, and then multiplies this volume change rate by the system's preset standard feed bulk density to calculate the real-time unloading mass flow rate. This measurement method completely encloses the detection area inside the truck compartment, isolating the detection process from external uneven ground and wind loads, thus improving the equipment's operational stability in complex environments.

[0003] However, bulk feed trucks often have to travel for extended periods on bumpy, unpaved roads in rural towns during their journey from feed mills to remote farms. Under the combined effects of continuous mechanical vibration and the feed's own weight, the granular or powdery feed inside the truck bed undergoes physical settling and compaction, resulting in a significant density gradient in the vertical direction. The bottom layer of feed is highly compacted, while the top layer is relatively loose. When unloading begins, the bottom auger prioritizes discharging the high-density feed from the bottom layer, while the top camera observes only the volume decrease of the top layer. When the system applies a uniform density parameter to the volume change of feed with a non-linear density gradient, a continuous deviation occurs between the calculated mass flow rate and the actual discharged physical mass. This reduces the accuracy of instantaneous unloading mass flow rate detection, thus affecting the precision of the intelligent agricultural product monitoring system in controlling the feed feeding process. Summary of the Invention

[0004] This application provides a method and system for detecting the unloading flow rate of bulk feed trucks based on visual recognition, which can improve the accuracy of unloading quality flow rate detection, thereby improving the precision of the intelligent monitoring system for agricultural products in the precise control of feed feeding procedures.

[0005] In the first aspect, this application provides a method for detecting the unloading flow rate of a bulk feed truck based on visual recognition, which calculates the reference density based on the initial total mass and initial total volume of the bulk feed truck when it is fully loaded.

[0006] Before the unloading mechanism is started, the initial spatial morphology of the feed surface inside the bulk feed truck compartment is obtained;

[0007] During the operation of the unloading mechanism, the real-time spatial morphology of the feed surface is acquired according to a preset time cycle;

[0008] The incremental discharge volume for the current cycle is calculated based on the change in the spatial boundary between the real-time spatial morphology of the current cycle and the real-time spatial morphology of the previous cycle.

[0009] Calculate the current cumulative discharge volume based on the spatial boundary difference between the current real-time spatial morphology and the initial spatial morphology;

[0010] Divide the current cumulative discharge volume by the initial total volume to obtain the unloading space evolution ratio;

[0011] According to the preset mapping relationship, the evolution ratio of the unloading space is converted into a density compensation coefficient, and the density compensation coefficient is positively correlated with the evolution ratio of the unloading space.

[0012] Multiply the reference density by the density compensation coefficient to obtain the instantaneous equivalent density for the current period;

[0013] Multiply the incremental discharge volume by the instantaneous equivalent density to obtain the incremental discharge mass for the current cycle;

[0014] The incremental unloading mass is accumulated to obtain the real-time cumulative unloading total mass.

[0015] By employing the above technical solution, the real-time spatial morphology of the feed surface is acquired at preset time intervals during the unloading process. Combined with the initial morphology, the incremental discharge volume and cumulative discharge volume are calculated, thus obtaining the unloading space evolution ratio. Since bulk feed is compacted at lower levels due to its own weight within the truck bed, resulting in differences in feed density at different depths, the unloading space evolution ratio is converted into a density compensation coefficient that is positively correlated with it. This coefficient dynamically reflects the changing degree of feed compaction as the unloading depth increases. Using this density compensation coefficient to correct the baseline density in real time yields the instantaneous equivalent density, making the volume-to-mass conversion process more consistent with the actual physical characteristics of bulk materials. This improves the accuracy of calculating the incremental unloading mass, ultimately enhancing the detection accuracy of the unloading flow rate and cumulative total unloading mass of bulk feed trucks.

[0016] In conjunction with some implementation methods of the first aspect, in some implementation methods, the unloading space evolution ratio is converted into a density compensation coefficient according to a preset mapping relationship, specifically including:

[0017] Map the evolution ratio of the unloading space to the relative depth feature of the current unloading interface;

[0018] Based on the preset distribution law of bulk lateral pressure, the local compressive stress corresponding to the relative depth characteristics is calculated. The local compressive stress is used to characterize the compaction effect of the feed weight above on the current depth.

[0019] Based on the preset stress density conversion model, the local extrusion stress is converted into the compaction deformation rate;

[0020] The density compensation coefficient is obtained by correcting the preset benchmark compensation coefficient based on the compaction deformation rate.

[0021] By adopting the above technical solution, the evolution ratio of the unloading space is mapped to a relative depth characteristic, and the local compressive stress is calculated in conjunction with the distribution law of bulk lateral pressure. This allows for a quantitative characterization of the actual compaction effect of the feed's own weight on the feed at the current depth. Furthermore, a stress-density conversion model is used to convert the local compressive stress into a compaction deformation rate, which is then used to correct the benchmark compensation coefficient. This ensures that the density compensation coefficient is obtained based on rigorous bulk mechanics principles, rather than simply through empirical mapping. This more realistically reflects the nonlinear density variation of feed at different depths, improving the rationality and reliability of the density compensation coefficient, and ultimately enhancing the accuracy of instantaneous equivalent density calculation.

[0022] In conjunction with some implementation methods of the first aspect, in some implementation methods, the unloading space evolution ratio is converted into a density compensation coefficient according to a preset mapping relationship, specifically including:

[0023] Extract the surface geometric features of the real-time spatial morphology in the current period;

[0024] The spatial dispersion of surface geometric features is calculated to obtain the morphological undulation index, which is used to characterize the degree of ruggedness of the feed surface caused by local flow during the unloading process.

[0025] The dynamic loosening coefficient of the current unloading interface is determined based on the difference between the morphological undulation index and the preset flatness benchmark value.

[0026] The density compensation coefficient is obtained by combining the preset basic compensation value corresponding to the evolution ratio of the unloading space with the dynamic loosening coefficient.

[0027] By employing the aforementioned technical solution, a morphological undulation index is obtained to characterize the degree of ruggedness on the feed surface caused by local flow by extracting the surface geometric features of the real-time spatial morphology and calculating the spatial dispersion. During the unloading process of bulk materials, the surface ruggedness is often accompanied by relative slippage and structural loosening between material particles, leading to a decrease in local actual density. The dynamic loosening coefficient is determined based on the difference between the morphological undulation index and the flat baseline value, and then fused with the basic compensation value. This allows for real-time capture and quantification of the weakening effect of the unloading dynamic process on feed density, avoiding overestimation of local density under conditions of intense material flow. This improves the dynamic adaptability of the density compensation mechanism under complex unloading conditions and further enhances the detection accuracy of unloading quality.

[0028] In conjunction with some embodiments of the first aspect, in some embodiments, after acquiring the real-time spatial morphology of the feed surface at preset time intervals during the operation of the unloading mechanism, the method further includes:

[0029] Extracting the spatial dynamics of real-time spatial morphology in continuous time series;

[0030] Based on the characteristics of spatial dynamic changes, identify the suspended occlusion areas in the real-time spatial morphology. The suspended occlusion areas represent the visual distortion of non-physical surfaces caused by unloading dust.

[0031] By stripping away the suspended occlusion areas in the real-time spatial morphology, discrete real material surface morphology is obtained.

[0032] Based on the boundary continuity of the actual material surface morphology, the spatial contour of the missing area caused by stripping is reconstructed to obtain the corrected target spatial morphology.

[0033] The target spatial shape is used as the real-time spatial shape of the current cycle to perform the step of calculating the incremental discharge volume of the current cycle based on the spatial boundary changes between the real-time spatial shape of the current cycle and the real-time spatial shape of the previous cycle.

[0034] By employing the aforementioned technical solution, and extracting the spatial dynamics of the real-time spatial morphology over a continuous time series, it is possible to effectively identify and remove suspended obstruction areas caused by unloading dust. Since dust produces visual distortions of non-solid surfaces in visual sensors, directly using morphological data containing distortion can lead to significant errors in volume calculations. By removing the suspended obstruction areas, discrete real material surface morphology is obtained, and spatial contour reconstruction of the missing areas is performed based on the boundary continuity of the real material surface morphology. This maximizes the restoration of the true physical contours of the dust-affected areas, reduces the interference of dust on visual detection results under harsh unloading conditions, improves the integrity and realism of the target spatial morphology, and thus enhances the robustness and accuracy of incremental discharge volume calculations.

[0035] In conjunction with some implementation methods of the first aspect, in some implementation methods, based on the characteristics of spatial dynamic changes, floating occlusion regions in the real-time spatial morphology are identified, specifically including:

[0036] From the characteristics of spatial dynamic changes, the frequency of depth jumps and spatial motion vectors of various local regions in the real-time spatial morphology are analyzed;

[0037] Local regions with a depth jump frequency greater than a preset frequency threshold are selected as candidate occlusion regions;

[0038] Calculate the corresponding directional dispersion based on the spatial motion vectors within the candidate occlusion region;

[0039] Candidate occlusion regions with directional dispersion greater than a preset dispersion threshold are identified as floating occlusion regions.

[0040] By employing the above technical solution, suspended obstructions such as unloading dust typically appear as violently fluctuating, high-frequency variations in depth information and chaotic movement directions under visual sensors, while the actual feed surface settles relatively smoothly and exhibits a consistent movement direction. By combining depth jump frequency and spatial motion vector direction dispersion for dual feature quantization screening, the system can effectively distinguish between actual feed surface morphological changes and visual distortions caused by dust. This reduces the likelihood of misjudging the actual feed surface as an obstruction area or missing dust, thereby improving the accuracy of suspended obstruction area identification and reducing errors caused by non-physical interference such as dust in feed surface morphology extraction.

[0041] In conjunction with some implementations of the first aspect, in some implementations, before reconstructing the spatial contour of the missing area resulting from peeling based on the boundary continuity of the actual material surface morphology to obtain the corrected target spatial morphology, the method further includes:

[0042] Extract the spatial displacement vector of the discrete real material surface morphology between the current cycle and the previous cycle to construct a bulk motion flow field that characterizes the slippage trend of the feed surface;

[0043] By extrapolating the kinematic flow field of the bulk material from the boundary of the actual material surface shape to the interior of the missing region, the internal deformation field of the missing region in the current cycle is predicted.

[0044] The internal deformation field is applied to the historical spatial morphology of the corresponding region in the previous cycle, and the physical deformation prior contour of the missing region is generated.

[0045] Using the physical deformation prior contour as the reference shape, the spatial contour reconstruction of the missing area generated by the stripping is performed based on the boundary continuity of the actual material surface shape.

[0046] By employing the aforementioned technical solution, a bulk material motion flow field is constructed by extracting the spatial displacement vectors of the actual material surface morphology between adjacent cycles. This flow field is then kinematically extrapolated from the boundary into the missing region to predict the internal deformation field. This internal deformation field is then applied to the historical spatial morphology of the corresponding region in the previous cycle, evolving to generate a priori physical deformation profile as a reconstruction benchmark. The flow of feed during unloading exhibits physical continuity and kinetic inertia. Extrapolating the displacement vectors of the actual material surface into the missing region can reasonably simulate the actual motion trend of the bulk material within the occluded area. Combining the predicted motion trend with the historical morphology to generate the priori profile ensures that the morphological reconstruction of the missing region is no longer limited to the static boundary geometric interpolation of the current cycle, but incorporates time-series evolution information consistent with the dynamics of bulk materials. This mechanism, combining historical morphology with kinematic prediction, makes the reconstructed spatial profile more closely match the actual physical deformation process of the feed surface, thereby improving the fidelity and reliability of the spatial profile reconstruction of the missing region.

[0047] In conjunction with some implementations of the first aspect, in some implementations, the flow field of the granular motion is kinematically extrapolated from the boundary of the actual material surface morphology to the interior of the missing region to predict the internal deformation field of the missing region in the current cycle, specifically including:

[0048] Extract the edge displacement vector at the boundary between the actual material surface shape and the missing region, and use it as the boundary condition for flow field extrapolation;

[0049] Based on the assumptions of mass conservation and flow continuity of granular materials, a Poisson equation constrained by the edge displacement vector is constructed.

[0050] Solving the Poisson equation yields the predicted displacement vectors of each spatial node within the missing region, and all predicted displacement vectors constitute the internal deformation field of the missing region in the current period.

[0051] By adopting the above technical solution, the flow of bulk feed follows specific laws of mass conservation and continuity on a macroscopic level. Using the actual edge displacement at the interface as boundary conditions and combining this with the aforementioned physical assumptions to construct the Poisson equation, the complex nonlinear motion of bulk feed can be transformed into a mathematical model strictly constrained by physical laws. Solving the Poisson equation to obtain the displacement vectors of internal nodes ensures that the prediction process of the internal deformation field is rigorously limited by physical conservation laws. This avoids the divergence or distortion of motion vectors caused by simply relying on geometric extrapolation, making the predicted internal deformation field more consistent with the actual flow characteristics of bulk materials, thereby improving the accuracy of the predicted deformation field in missing regions.

[0052] Secondly, embodiments of this application provide a visual recognition-based bulk feed truck unloading flow detection system, which includes: one or more processors and a memory; the memory is coupled to one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and one or more processors call the computer instructions to cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0053] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a system, cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0054] Fourthly, embodiments of this application provide a computer program product that, when run on a system, causes the system to execute the method described in any possible implementation of the first aspect.

[0055] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0056] 1. This application provides a visual recognition-based method for detecting the unloading flow rate of bulk feed trucks. By acquiring the real-time spatial morphology of the feed surface at preset time intervals during the unloading process, and combining this with the initial morphology to calculate the incremental discharge volume and cumulative discharge volume, the unloading space evolution ratio is obtained. Since bulk feed is compacted at lower levels due to its own weight within the truck bed, resulting in differences in feed density at different depths, the unloading space evolution ratio is converted into a density compensation coefficient that is positively correlated with this density. This coefficient dynamically reflects the changing degree of feed compaction as the unloading depth increases. Using this density compensation coefficient to correct the baseline density in real time yields the instantaneous equivalent density, making the volume-to-mass conversion process more consistent with the actual physical characteristics of bulk materials. This improves the accuracy of calculating the incremental unloading mass, ultimately enhancing the detection accuracy of the unloading flow rate and cumulative unloading mass of the bulk feed truck.

[0057] 2. This application provides a visual recognition-based method for detecting the unloading flow rate of bulk feed trucks. By extracting the spatial dynamic change features of the real-time spatial morphology in a continuous time series, it can effectively identify and remove suspended obstruction areas caused by unloading dust. Since dust produces visual distortions of non-solid surfaces in visual sensors, directly using morphological data containing distortions can lead to significant errors in volume calculation. By removing the suspended obstruction areas, discrete real material surface morphology is obtained, and spatial contour reconstruction of the missing areas is performed based on the boundary continuity of the real material surface morphology. This maximizes the restoration of the true physical contour of the dust-affected areas, reduces the interference of dust on visual detection results under harsh unloading conditions, improves the integrity and realism of the target spatial morphology, and thus enhances the robustness and accuracy of incremental discharge volume calculation.

[0058] 3. This application provides a visual recognition-based method for detecting the unloading flow rate of bulk feed trucks. It constructs a bulk material motion flow field by extracting the spatial displacement vectors of the actual material surface morphology between adjacent cycles. This flow field is then kinematically extrapolated from the boundary into the missing region to predict the internal deformation field. The internal deformation field is then applied to the historical spatial morphology of the corresponding region in the previous cycle, evolving to generate a priori physical deformation profile as a reconstruction benchmark. The flow of feed during unloading exhibits physical continuity and motion inertia. Extrapolating the displacement vectors of the actual material surface into the missing region can reasonably simulate the actual motion trend of the bulk material within the obstructed area. Combining the predicted motion trend with the historical morphology to generate the priori profile ensures that the morphology reconstruction of the missing region is no longer limited to the static boundary geometric interpolation of the current cycle, but incorporates time-series evolution information consistent with the dynamics of the bulk material. This mechanism, combining historical morphology and kinematic prediction, makes the reconstructed spatial profile more closely match the actual physical deformation process of the feed surface, thereby improving the fidelity and reliability of the spatial profile reconstruction of the missing region. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating a method for detecting the unloading flow rate of a bulk feed truck based on visual recognition, as described in an embodiment of this application.

[0060] Figure 2 This is another flowchart illustrating a method for detecting the unloading flow rate of a bulk feed truck based on visual recognition, as described in this application.

[0061] Figure 3 This is another flowchart illustrating a method for detecting the unloading flow rate of bulk feed trucks based on visual recognition, as described in this application.

[0062] Figure 4 This is a schematic diagram of the physical device structure of a bulk feed truck unloading flow detection system based on visual recognition, provided in an embodiment of this application. Detailed Implementation

[0063] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0064] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0065] The following example is used in conjunction with Figure 1 The present application describes a method for detecting the unloading flow rate of bulk feed trucks based on visual recognition in its embodiments:

[0066] Please see Figure 1 This is a flowchart illustrating a method for detecting the unloading flow rate of a bulk feed truck based on visual recognition, as described in an embodiment of this application.

[0067] S101. The baseline density is calculated based on the initial total mass and initial total volume of the bulk feed truck when it is fully loaded.

[0068] Initial gross mass refers to the total weight of the feed loaded inside the bulk feed truck before unloading begins. This weight can be obtained from various sources, including but not limited to weighbridge data, on-board weighing sensor data, or factory delivery records. Initial gross volume refers to the actual three-dimensional physical space occupied by the feed inside the truck bed under the same full-load condition. This is not limited to the volume of pelleted, powdered, or mixed feed. Reference density refers to the average physical density obtained by completely distributing the initial gross mass evenly across the initial gross volume at a macroscopic level. This density is a static reference value and does not consider density differences between layers inside the truck bed due to gravity compaction.

[0069] The system can interact with an external IoT weighing platform to obtain the weighing record of the bulk feed truck before it enters the unloading area by calling the application programming interface. It extracts the gross weight and subtracts the tare weight of the corresponding license plate number pre-stored in the vehicle database to accurately calculate the initial total mass. When obtaining the initial total volume, the system controls a multi-line LiDAR deployed on the top of the truck body to perform a full-coverage scan of the interior of the truck body to obtain high-precision three-dimensional point cloud data. After filtering and denoising the point cloud data, the system extracts the contour coordinates of the feed surface and, combined with a pre-imported standard three-dimensional digital model of the bulk feed truck body, uses a mesh generation and volume integration algorithm to calculate the initial total volume of the envelope between the point cloud surface and the bottom surface of the truck body. Finally, the system divides the initial total mass by the initial total volume to obtain the baseline density and stores it in memory.

[0070] The system can also directly extract the initial total mass recorded in the accompanying electronic cargo manifest by parsing the message data sent by the on-board terminal of the bulk feed truck. For the calculation of the initial total volume, the system drives a depth camera array installed inside the truck body to collect multi-view depth images under full load conditions, uses image processing algorithms to extract feature points on the feed surface, and stitches the multi-view depth data into a complete feed surface depth map through coordinate system transformation. The system calculates the physical height corresponding to each pixel point based on the pixel value in the depth map, and calculates the initial total volume by using the bottom area of ​​the truck body as a reference and the pixel cylinder accumulation method. Then the system executes an arithmetic division instruction to divide the mass data by the volume data to generate a baseline density value for subsequent process calls.

[0071] S102. Before the unloading mechanism is started, obtain the initial spatial shape of the feed surface inside the bulk feed truck compartment;

[0072] The unloading mechanism refers to the mechanical or pneumatic device on a bulk feed truck used to transport feed from inside the truck bed to an external storage tower or trough, including but not limited to augers, pneumatic conveying pipelines, or conveyor belt systems. The initial spatial morphology refers to the geometric shape, elevation distribution, and boundary contour features of the top surface of the feed layer inside the truck bed in three-dimensional space before the unloading action occurs; its representation is not limited to a three-dimensional point cloud model, depth image matrix, or elevation grid map.

[0073] The system can acquire left and right view images inside the truck bed before the control circuit of the unloading mechanism motor is closed by sending a trigger command to an industrial-grade binocular stereo vision camera installed on the top of the truck bed. The system uses a stereo matching algorithm to perform epipolar correction and disparity calculation on the left and right views to generate a high-resolution disparity map. Then, based on the camera's intrinsic and extrinsic parameter matrix, the disparity map is converted into three-dimensional point cloud data. The system performs statistical filtering and voxel downsampling on these point cloud data to remove noise caused by suspended dust, and finally constructs a smooth and continuous three-dimensional mesh model of the feed surface, which is used as the initial spatial form for persistent storage.

[0074] The system can also project a specially coded infrared spot pattern onto the feed surface using a structured light scanner inside the carriage, and capture images of the deformed spot using an infrared camera. The system's built-in decoding algorithm calculates the depth information of each discrete point on the feed surface based on the degree of deformation of the spot, generating an elevation matrix containing three-dimensional coordinates. The system further performs interpolation and smoothing on the elevation matrix to fill in local data gaps caused by reflection or occlusion, thereby generating a complete elevation topology map of the feed surface. This map is defined as the initial spatial morphology and stored in the system's baseline database, providing an initial data source for subsequent volume difference calculations.

[0075] S103. During the operation of the unloading mechanism, the real-time spatial morphology of the feed surface is acquired according to a preset time cycle;

[0076] The preset time period refers to the time interval between two adjacent data acquisition actions during continuous monitoring. The setting of this period is not limited to the rated power of the unloading mechanism, the estimated unloading flow rate, or the maximum sampling frame rate of the vision sensor. Real-time spatial morphology refers to the transient three-dimensional geometric shape and elevation distribution of the feed surface at a specific moment during the continuous operation of the unloading mechanism and the continuous discharge of feed from the truck. Its data structure is not limited to point cloud sets, depth maps, or three-dimensional surface equations.

[0077] The system can set a fixed millisecond-level interrupt period through an internal hardware timer module. Whenever the timer overflows and triggers an interrupt, the system sends a synchronization acquisition signal to the time-of-flight ranging camera deployed above the carriage. The time-of-flight ranging camera obtains the depth matrix of the feed surface at the current moment by emitting modulated infrared light and receiving reflected light, and calculating the photon flight time. The system binds the received depth matrix with the current timestamp and uses a median filtering algorithm to eliminate random interference caused by dust raised during unloading on the depth measurement, thereby generating real-time spatial morphology data for the current period and pushing it into the system's data processing queue.

[0078] The system can also dynamically adjust the preset time period based on the real-time operating speed of the unloading mechanism through a multi-threaded polling mechanism at the software level. When the speed is faster, the period is shortened, and when the speed is slower, the period is extended. At each dynamically calculated period node, the system calls the laser contour scanner in the carriage to control the laser beam to perform a rapid lateral scan on the feed surface and obtain multiple two-dimensional contour lines. The system combines the longitudinal movement step or rotation angle of the scanner to stitch these two-dimensional contour lines together and reconstruct the three-dimensional surface topology at the current moment. After coordinate system alignment and surface smoothing, it is output as the real-time spatial form of the current period to the subsequent volume calculation module.

[0079] S104. Calculate the incremental discharge volume of the current cycle based on the spatial boundary change between the real-time spatial form of the current cycle and the real-time spatial form of the previous cycle.

[0080] Spatial boundary change refers to the difference in physical displacement and deformation of the feed surface's contour, height, and surface morphology in three-dimensional space between two adjacent sampling periods. Its representation is not limited to height difference matrices, point cloud distance fields, or volume Boolean operation results. Incremental discharge volume refers to the physical size of the feed actually transferred from inside the truck bed to the outside within a preset time period due to the suction or pushing action of the unloading mechanism. It is not limited to the sum of the volumes of infinitesimal elements with a specific geometric shape.

[0081] The system can convert the real-time spatial morphology of the current cycle and the real-time spatial morphology of the previous cycle into an elevation grid model under the same coordinate system. The system traverses the grid nodes with the same planar coordinates in the two grid models, performs subtraction to obtain the elevation difference at each node, and for the region where the elevation decreases, the system multiplies the elevation difference by the fixed planar area of ​​a single grid to obtain the micro-volume of the local region. Then, the system sums up all the micro-volumes, removes abnormally negative volumes caused by measurement errors, and finally obtains the incremental discharge volume for the current cycle.

[0082] The system can also be implemented using 3D point cloud processing technology. The system extracts 3D point cloud data of the real-time spatial morphology of the current cycle and the previous cycle respectively, and uses the nearest point iteration algorithm to ensure accurate registration of the two sets of point clouds in the spatial coordinate system. Then, the system uses the point cloud convex hull algorithm or Delaunay triangulation algorithm to construct closed surfaces of the point cloud data of the two cycles respectively. The system calls the 3D geometry calculation engine to perform Boolean subtraction on the two closed surfaces and directly calculates the volume of the closed space between the two surfaces. The system uses the value of the closed space volume as the incremental volume of the current cycle and stores it in the system's temporary register for subsequent mass conversion.

[0083] S105. Calculate the current cumulative discharge volume based on the spatial boundary difference between the real-time spatial form and the initial spatial form of the current cycle.

[0084] Spatial boundary difference refers to the cumulative elevation drop and contour evolution between the feed surface morphology at the current monitoring moment and the absolute initial morphology before the unloading action, within the entire three-dimensional spatial coordinate system. Its manifestation is not limited to a global height difference map, cumulative deformation field, or macroscopic volume difference. The current cumulative discharge volume refers to the total reduction in physical feed space within the bulk feed truck compartment from the initial start of the unloading mechanism until the moment the real-time spatial morphology is acquired. It is not limited to a simple mathematical summation of the incremental discharge volumes across different cycles; it can also be a direct calculation based on the global morphology.

[0085] The system can calculate the discharge volume by establishing a global cylindrical integral model. The system projects the initial spatial shape and the real-time spatial shape of the current cycle onto a two-dimensional plane on the bottom of the carriage, dividing it into a fine two-dimensional pixel array. For each pixel coordinate in the array, the system reads the depth value from the initial shape depth map and the current shape depth map respectively. The system calculates the absolute difference between these two depth values, which is the total height of feed descent at that coordinate point. The system multiplies this total height by the actual physical area corresponding to the pixel to obtain the discharge volume of a single pixel cylinder. Finally, the system traverses the entire two-dimensional pixel array, integrates and sums the discharge volumes of all pixel cylinders, thereby calculating the current cumulative discharge volume.

[0086] The system can also be implemented by constructing a three-dimensional voxel mesh space. The system divides the effective loading space inside the carriage into uniform three-dimensional voxel blocks. Based on the three-dimensional surface equation of the initial spatial shape, the system marks the set of voxels occupied by feed in the initial state. Then, based on the three-dimensional surface equation of the real-time spatial shape in the current cycle, the system marks the set of voxels still occupied by feed in the current state. The system performs a difference operation on these two sets to count the total number of voxels whose state changes from occupied to free. The system multiplies the total number of voxels by the known physical volume of a single standard voxel to obtain the current cumulative discharge volume and updates this value to the main control panel display area of ​​the system.

[0087] S106. Divide the current cumulative discharge volume by the initial total volume to obtain the unloading space evolution ratio;

[0088] The unloading space evolution ratio refers to the relative ratio of the volume of feed that has been discharged to the total volume of feed when fully loaded. This ratio is a dimensionless parameter used to macroscopically quantify the progress and depth of the current unloading process. Its value range is usually between zero and one or presented as a percentage, and it is not limited to linear progress indicators or space emptying rate.

[0089] The system can directly read the current cumulative discharge volume stored in the memory address and the initial total volume value stored in the read-only memory during system initialization by calling the floating-point unit of the central processing unit. The system executes a standard division instruction to calculate a decimal between zero and one. The system further performs low-pass filtering on the decimal to smooth out the small fluctuations caused by sensor measurement noise, ensuring the monotonically increasing characteristic of the ratio value. Finally, the processed decimal is defined as the unloading space evolution ratio and stored in the system's global state variables.

[0090] The system can also be implemented by constructing a state machine model of the unloading progress. The system divides the initial total volume into one hundred equal intervals. The system receives the data stream of the current cumulative discharge volume in real time, uses comparator logic to determine which equal interval the current cumulative discharge volume falls into, and calculates a percentage value based on the matched interval index value. To improve accuracy, the system uses a linear interpolation algorithm within the interval to calculate the specific floating-point percentage based on the distance between the current cumulative volume and the interval boundary. The system uses this percentage value with decimal places as the unloading space evolution ratio and broadcasts it to all submodules that need to perform density compensation calculations through the internal bus.

[0091] S107. Based on the preset mapping relationship, convert the unloading space evolution ratio into a density compensation coefficient;

[0092] The system converts the unloading space evolution ratio into a density compensation coefficient based on a preset mapping relationship. The density compensation coefficient is positively correlated with the unloading space evolution ratio. The preset mapping relationship refers to a rule or model pre-established within the system to describe the mathematical relationship between unloading progress and feed density changes; it is not limited to linear functions, nonlinear curve fitting equations, or multidimensional lookup tables. The density compensation coefficient is a multiplier factor used to correct the baseline density; its value reflects the degree of deviation of the actual feed density from the baseline density at the current unloading depth. Relative depth characteristics refer to the relative vertical position of the current unloading interface within the overall truck bed height. Local compressive stress refers to the internal compressive pressure generated by the gravity of the remaining material above the feed particles at a specific depth. Compaction deformation rate refers to the proportion of volumetric compression deformation of feed particles under local compressive stress. The baseline compensation coefficient refers to the initial density correction value considering only the depth factor under ideal flat conditions. Surface geometric characteristics refer to the three-dimensional topological structure attributes of the current unloading interface. Spatial dispersion refers to the statistical dispersion of surface elevation data from the average elevation. The morphological undulation index is a specific numerical value quantifying the degree of surface roughness in feed. The dynamic looseness coefficient is a density reduction factor caused by surface loosening due to localized feed flow and collapse. The flatness benchmark value is the threshold value of the morphological undulation index for an ideally perfectly smooth surface. This threshold is derived from extensive calibration experimental data on standard flat material surfaces; the greater the difference between the morphological undulation index and the flatness benchmark value, the more uneven the surface. The preset basic compensation value is the initial compensation amount obtained solely based on the unloading ratio. This step can be achieved in at least two ways:

[0093] The system can map the evolution ratio of the unloading space to the relative depth characteristics of the current unloading interface; calculate the local compressive stress corresponding to the relative depth characteristics according to the preset bulk lateral pressure distribution law, and the local compressive stress is used to characterize the compaction effect of the feed weight above on the current depth; convert the local compressive stress into the compaction deformation rate according to the preset stress density conversion model; and correct the preset benchmark compensation coefficient based on the compaction deformation rate to obtain the density compensation coefficient. The system can directly substitute the evolution ratio of the unloading space into the preset depth conversion equation to map the relative depth characteristics of the current unloading interface. Based on the preset bulk lateral pressure distribution law, the system inputs the relative depth characteristics and the internal friction angle parameters of the feed to calculate the local compressive stress at the current depth. Next, the system utilizes a preset nonlinear stress density conversion model, which is a deep neural network built on porous media compression theory. Because the neural network has been fully trained using a large amount of feed compression experimental data, it can accurately fit the nonlinear stress-deformation relationship. The system inputs the local compressive stress into the model, and the model outputs the corresponding compaction deformation rate through forward propagation. The system extracts the built-in benchmark compensation coefficient and uses the calculated compaction deformation rate as a weighting parameter to perform a weighted correction operation on the benchmark compensation coefficient, thereby obtaining the final density compensation coefficient.

[0094] The system can also extract the surface geometric features of the real-time spatial morphology of the current cycle; calculate the spatial dispersion of the surface geometric features to obtain the morphological undulation index, which is used to characterize the degree of ruggedness of the feed surface caused by local flow during the unloading process; determine the dynamic loosening coefficient of the current unloading interface based on the difference between the morphological undulation index and the preset flatness benchmark value; and calculate the density compensation coefficient by fusing the preset basic compensation value corresponding to the unloading space evolution ratio with the dynamic loosening coefficient. The system can also extract the surface geometric features of the real-time spatial morphology of the current period. The system uses principal component analysis algorithm to process the surface point cloud and extract the normal vector distribution matrix as the surface geometric features. Then, the system calculates the variance of the normal vector distribution matrix as the spatial dispersion, thereby obtaining the morphological undulation index. The system subtracts the morphological undulation index from the flatness benchmark value stored in the system configuration file, and calls the preset attenuation function according to the difference to determine the dynamic loosening coefficient of the current unloading interface. Finally, the system queries the preset one-dimensional data table to obtain the preset basic compensation value according to the unloading space evolution ratio. The system multiplies the preset basic compensation value with the dynamic loosening coefficient and performs a fusion calculation to finally obtain the accurate density compensation coefficient.

[0095] S108. Multiply the reference density by the density compensation coefficient to obtain the instantaneous equivalent density of the current period.

[0096] Instantaneous equivalent density refers to a dynamic value that truly represents the actual physical density of the feed being discharged within a specific unloading time period, after being corrected by the combined effects of deep compaction and surface loosening. It is not limited to a single scalar value and can also be a density matrix for different unloading areas.

[0097] The system can call its internal digital signal processor to read the baseline density data calculated and saved during the initialization phase from the static random access memory. At the same time, it retrieves the density compensation coefficient for the current period, which was just calculated by the pre-compensation module, from the cache queue. The system uses a hardware multiplier to execute high-precision floating-point multiplication instructions to multiply the baseline density by the density compensation coefficient. The system performs effective truncation and range verification on the result of the multiplication operation to prevent density values ​​from exceeding the limit due to abnormal data. After the verification is passed, the system uses the product result as the instantaneous equivalent density for the current period and writes it to a dedicated shared memory area for subsequent mass calculation modules to read.

[0098] The system can also be implemented by building a software-based streaming data processing pipeline. The system configures the baseline density as a constant node in the data flow graph and the real-time generated density compensation coefficient as a dynamic input node. The system defines a multiplication operator node in the data flow graph. When the dynamic input node receives a new density compensation coefficient, the system automatically triggers the multiplication operator node to perform the operation. The system multiplies the value of the constant node with the value of the dynamic node. The system can also introduce a small ambient temperature correction factor to participate in the multiplication operation to compensate for the slight influence of temperature on feed density. Finally, the system outputs the instantaneous equivalent density of the current period and pushes the density value to the next calculation stage through a message queue.

[0099] S109. Multiply the incremental discharge volume by the instantaneous equivalent density to obtain the incremental discharge mass for the current cycle;

[0100] Incremental unloading mass refers to the physical weight of feed actually discharged from the bulk feed truck within the current preset time period. It is the final result of the combined effect of volume and density, and its unit of measurement is not limited to kilograms, tons, or pounds.

[0101] The system can synchronously acquire the incremental discharge volume value of the current cycle broadcast on the system bus and the instantaneous equivalent density value read from shared memory through the arithmetic logic unit of the main control program. The system performs double-precision floating-point multiplication, multiplying the volume value and the density value. In order to eliminate the small rounding errors that may occur during the calculation, the system uses the banker's rounding method to control the precision of the product result, retaining three significant digits after the decimal point. The system defines the processed value as the incremental unloading mass of the current cycle and encapsulates it into a comprehensive data packet containing timestamp, volume, density and mass, and sends it to the system's data recording module for archiving.

[0102] The system can also implement this step by constructing a hardware acceleration module based on a field-programmable gate array (FPGA). The system converts the incremental discharge volume and instantaneous equivalent density into a fixed-point number format and inputs it into the FPGA chip. The chip is pre-configured with a high-speed parallel multiplier logic core. The system uses this logic core to quickly complete the multiplication operation of volume and density within one clock cycle. The system then converts the calculated fixed-point number result back into a floating-point number format to obtain the incremental discharge mass of the current cycle. The system transmits this mass value to the display terminal in real time through a serial peripheral interface, displaying the changing trend of the current instantaneous discharge flow rate in the form of a dynamic data point graph.

[0103] S110. The incremental unloading mass is accumulated to obtain the real-time cumulative unloading total mass.

[0104] The real-time cumulative total unloading mass refers to the total physical weight of all feed that the bulk feed truck has transported from the moment the unloading operation officially begins until the end of the current monitoring cycle. It is not limited to the simple historical incremental accumulation value, but can also include the dynamic smoothing correction amount of the system for historical errors.

[0105] The system can initialize a double-precision floating-point global accumulator variable in memory, with an initial value of zero. After each preset time period ends and the incremental unloading mass of the current period is successfully calculated, the system calls the atomic addition operation instruction to safely accumulate the value of the incremental unloading mass into the global accumulator variable. The system uses a mutex lock mechanism to ensure the thread safety of the accumulation operation in a multi-threaded environment and prevent data read and write conflicts. After the accumulation is completed, the system reads the latest value of the global accumulator variable, which is the real-time accumulated total unloading mass. The system pushes this total mass value to the main display area of ​​the vehicle human-machine interface for real-time large-font refresh display.

[0106] The system can also accumulate and manage data by building a time-series database. The system treats the incremental unloading mass calculated in each period as an independent data record, with an accurate current timestamp, and continuously inserts it into a specific table in the time-series database. The system runs an aggregation query engine in the background. Whenever new data is inserted, the engine automatically triggers a summation aggregation function to sum all incremental unloading mass records in the table after the unloading start time. The system obtains the result of this aggregation query as the real-time cumulative total unloading mass. At the same time, the system uses the data in the database to generate an integral curve of unloading mass changing over time in real time, and synchronously uploads the cumulative total mass and curve data to the cloud server via wireless network for real-time viewing by the remote monitoring center.

[0107] In the above embodiments, the real-time spatial morphology of the feed surface is acquired at preset time intervals during the unloading process, and the incremental discharge volume and cumulative discharge volume are calculated in combination with the initial morphology to obtain the unloading space evolution ratio. Since bulk feed is compacted at the bottom due to its own weight within the truck bed, resulting in differences in feed density at different depths, the unloading space evolution ratio is converted into a density compensation coefficient that is positively correlated with it. This dynamically reflects the degree of feed compaction as the unloading depth increases. Using this density compensation coefficient to correct the baseline density in real time yields the instantaneous equivalent density, making the volume-to-mass conversion process more consistent with the actual physical characteristics of bulk materials. This improves the accuracy of calculating the incremental unloading mass, ultimately enhancing the detection accuracy of the unloading flow rate and cumulative total unloading mass of the bulk feed truck.

[0108] In the above embodiments, although accurate calculation of basic unloading flow rate and mass is achieved through density dynamic compensation, a large amount of dust is often generated during the actual unloading process of bulk feed. This dust forms suspended obstructions in the field of view of the visual sensor, causing visual distortion of non-solid surfaces in the acquired real-time spatial shape, which seriously affects the accuracy of subsequent volume calculation. To solve this interference problem under actual working conditions and further improve the detection accuracy of the system in harsh environments, this application embodiment introduces a shape correction mechanism after acquiring the real-time spatial shape. The following is combined with... Figure 2 Another method for detecting the unloading flow rate of bulk feed trucks based on visual recognition is described in the embodiments of this application:

[0109] Please see Figure 2 This is another flowchart illustrating a method for detecting the unloading flow rate of a bulk feed truck based on visual recognition, as described in this application.

[0110] S201. Extract the spatial dynamic change characteristics of real-time spatial morphology in continuous time series;

[0111] Real-time spatial morphology refers to the three-dimensional structural data of the surface of bulk feed acquired by a visual sensor at a specific moment. Its forms include, but are not limited to, three-dimensional point cloud data, depth image matrices, or three-dimensional mesh models. Continuous time series refers to a collection of data frames continuously acquired at certain time intervals, such as a video stream at a fixed frame rate or a continuous sequence of three-dimensional point clouds. Spatial dynamic change characteristics refer to the patterns and attributes of changes in the position or state of data points in space over time, including but not limited to spatial displacement vectors, depth fluctuation rates, or local velocity fields.

[0112] The system can convert real-time spatial morphology from continuous time series into a two-dimensional depth image sequence and utilize a dense optical flow estimation algorithm to extract spatial dynamic change features. Specifically, the system first applies Gaussian smoothing filtering to adjacent depth image frames to eliminate high-frequency sensor noise. Then, it uses edge detection operators to calculate the spatial gradients of the images in the horizontal and vertical directions, as well as the temporal gradients of pixels at the same location between adjacent frames. Based on the assumption of motion consistency between adjacent pixels, the system constructs optical flow constraint equations and calculates the motion velocity components of each pixel in the two-dimensional plane through iterative solutions. Finally, combining the camera's intrinsic parameter matrix and depth information, the two-dimensional velocity components are back-projected into three-dimensional space to generate a three-dimensional motion vector for each spatial point, which serves as the spatial dynamic change feature.

[0113] The system can also directly extract features from continuous time-series 3D point cloud data using scene flow statistical algorithms. First, the system divides the original point cloud into multiple uniform 3D voxel grids and extracts local geometric feature descriptors, such as fast point feature histograms, within each voxel. Next, the system performs nearest neighbor search between point clouds of adjacent frames using the similarity of the feature descriptors to establish 3D spatial correspondences for local regions. Based on these correspondences, the system calculates the 3D translation vector and rotation matrix of each local region from the previous frame to the current frame, thereby obtaining spatial motion vectors that accurately reflect the displacement of spatial points, which are then used as the extracted spatial dynamic change features.

[0114] S202. Identify suspended occlusion areas in the real-time spatial morphology based on the characteristics of spatial dynamic changes;

[0115] The system identifies suspended occlusion regions in the real-time spatial morphology based on dynamic spatial change characteristics. These regions represent visual distortions of non-physical surfaces caused by unloading dust. A local region refers to a smaller data unit obtained by spatially dividing the overall real-time spatial morphology, including but not limited to two-dimensional image blocks or three-dimensional voxel nodes. Depth jump frequency refers to the number of times the depth value of a local region undergoes a drastic change exceeding the normal noise range within a set time window. Spatial motion vector refers to the direction and distance of the data point within the local region's movement in three-dimensional space. The preset frequency threshold is a baseline value used to measure whether depth changes are abnormal; it is derived from the upper limit of statistical data on depth fluctuations caused by sensor hardware noise and normal material drop under ideal dust-free conditions. Candidate occlusion regions refer to areas initially screened as having abnormal depth changes and suspected of containing dust. Directional dispersion is a statistical measure of the consistency of the directions of multiple spatial motion vectors within a local region. The preset dispersion threshold is a baseline value used to distinguish between regular and disordered motion; it is derived from aerodynamic comparative analysis data of the directional motion trajectory of solid feed falling under gravity and the Brownian motion trajectory of dust affected by airflow. There is a cascaded filtering logic relationship between the preset frequency threshold and the preset discrete threshold. The former serves as the initial screening condition in the time dimension, used to quickly remove static backgrounds, while the latter serves as the fine screening condition in the spatial dimension, used to accurately distinguish between normal material falling and disorderly dust in dynamic areas.

[0116] Specifically, the system analyzes the depth jump frequency and spatial motion vector of each local region in the real-time spatial morphology from the dynamic changes in space. Local regions with a depth jump frequency greater than a preset frequency threshold are selected as candidate occlusion regions. Based on the spatial motion vectors within the candidate occlusion regions, the corresponding directional dispersion is calculated. Candidate occlusion regions with directional dispersion greater than a preset dispersion threshold are identified as floating occlusion regions. This refinement technique can be implemented using a statistical analysis method based on octree spatial partitioning. The system first constructs the real-time spatial morphology as an octree data structure, treating each leaf node as a local region. The system maintains a fixed-length time sliding window queue for each leaf node, recording the average depth across multiple consecutive frames. By calculating the absolute value of the depth difference between adjacent frames, the number of times it exceeds the environmental noise tolerance is counted to obtain the depth jump frequency. The system marks leaf nodes with a depth jump frequency greater than a preset frequency threshold as candidate occlusion regions. Subsequently, the system extracts the spatial motion vectors of all data points within the candidate occlusion regions, maps them onto a unit sphere, and calculates the variance of the distribution of these vector endpoints on the sphere as the directional dispersion. If the variance is greater than the preset discrete threshold, the system will determine the spatial range corresponding to the leaf node as the floating occlusion region.

[0117] The system can also utilize a 3D convolutional neural network based on spatiotemporal feature fusion to achieve this refinement technique. The system transforms the real-time spatial morphology into a four-dimensional voxel tensor containing a time dimension, dividing it into multiple spatiotemporal sub-blocks as local regions. The system calculates the number of depth abrupt change frames on the time axis of each spatiotemporal sub-block using a difference algorithm, using this as the depth jump frequency, and filters out spatiotemporal sub-blocks with frequencies exceeding a preset threshold as candidate occlusion regions. Next, the system inputs the spatial motion vector field within the candidate occlusion regions into a pre-trained 3D convolutional neural network. This model is able to calculate and identify directional dispersion because it inputs massive amounts of labeled motion vector field data of normal material falling and dust interference during the training phase. The multi-layered 3D convolutional kernels within the model continuously adjust their weights through a backpropagation algorithm, deeply learning and memorizing the nonlinear topological differences between the disordered eddy current characteristics in dust fluid dynamics and the parallel flow characteristics in the rigid body dynamics of solid falling materials. The model extracts features and reduces the dimensionality of the input motion vector field. Finally, it outputs a probability score representing the degree of motion disorder as the directional discreteness through a fully connected layer. The system identifies spatiotemporal sub-blocks with scores greater than a preset discrete threshold as floating occlusion regions.

[0118] S203. Strip away the suspended occlusion area in the real-time spatial form to obtain the discrete real material surface form;

[0119] Stripping refers to the process of completely removing, hiding, or marking as unusable specific invalid or interfering data from the original dataset through data structure operations or algorithmic filtering, including but not limited to point cloud cropping, mask overlay, or invalid value replacement. Discrete real material surface morphology refers to the remaining three-dimensional spatial data after removing dust interference data. Because the data of areas originally obscured by dust is cleared, the remaining surface data loses its integrity in spatial topology, presenting a discontinuous state with holes, faults, or fragmentation, including but not limited to sparse point cloud sets or depth maps with invalid pixel backgrounds.

[0120] The system employs a point cloud Boolean subtraction technique based on spatial bounding boxes to achieve the stripping process. First, the system acquires the extreme 3D coordinates of all identified suspended occlusion regions. Based on this, it constructs a series of tightly fitting axial or directed bounding boxes in 3D space. Then, the system traverses every 3D data point in the current real-time spatial morphology, using a point-polygon positional relationship algorithm to determine whether each data point is located inside any bounding box. For data points determined to be inside a bounding box, the system directly releases their memory address from the dynamic point cloud array in memory or marks their index as deleted. After the complete traversal and removal operation, the system repackages and outputs the remaining unmarked data points, resulting in a discrete, realistic material surface morphology with spatial voids, free from dust interference.

[0121] The system can also employ a technique based on ray tracing and depth buffer masks to achieve data stripping. The system simulates the physical imaging principle of a visual sensor, emitting dense tracing rays from the optical center of a virtual camera towards the real-time spatial morphology. When a ray intersects a data point in space, the system checks if the intersection point belongs to an identified floating occlusion region. If it does, the system forcibly modifies the depth value of the pixel location traversed by the ray in the corresponding two-dimensional depth buffer matrix to a special identifier representing infinity or invalidity. After tracing all rays, a binary mask precisely corresponding to the dust location is formed in the depth buffer matrix. Finally, the system regenerates the three-dimensional surface based on this depth buffer matrix with invalid identifiers. Areas covered by the mask will not generate any spatial points, thus achieving geometric data stripping and outputting a discrete, realistic material surface morphology exhibiting fragmented characteristics.

[0122] S204. Based on the boundary continuity of the actual material surface shape, the spatial contour of the missing area caused by stripping is reconstructed to obtain the corrected target spatial shape.

[0123] Boundary continuity refers to the surface geometry of bulk feed, when naturally piled up under gravity, following the law of the angle of repose in physics. This manifests as a smooth transition in surface height and slope, without any vertical cliffs or abrupt changes that violate physical laws. In other words, the effective data points at the edge of the missing region maintain consistency in mathematical gradient and curvature with the unknown real surface inside. The missing region refers to the blank space or topological void left in the discrete real feed surface after a stripping operation, where no data is covered. Spatial contour reconstruction refers to the data processing process of using known data and specific mathematical or physical models to calculate and generate three-dimensional coordinate points inside the missing region to fill the void. This includes, but is not limited to, surface fitting, spatial interpolation, or learning-based morphology generation. The target spatial morphology refers to the three-dimensional data model that, after reconstruction and repair, eliminates data voids, restores overall topological integrity, and accurately reflects the real physical contour of the feed pile at the current moment.

[0124] The system employs a local surface fitting algorithm based on moving least squares to reconstruct spatial contours. First, the system uses an edge detection algorithm to identify the boundary data points of each missing region within the discrete real material surface morphology. For each target mesh coordinate that needs to be reconstructed within the missing region, the system sets a search radius centered on that coordinate and obtains all valid boundary data points within that radius as control points. Next, the system assigns weights to these control points, with control points closer to the target coordinates having higher weights. The system constructs a quadratic or cubic polynomial surface equation and uses the moving least squares method to solve the equation coefficients, ensuring that the surface best fits these control points in a weighted sense. After solving, the target mesh coordinates are substituted into the surface equation to calculate the corresponding depth value. After point-by-point filling, a smooth and continuous target spatial morphology is generated.

[0125] The system can also utilize a 3D point cloud completion model based on generative adversarial networks (GANs) to reconstruct spatial contours. The system inputs discrete, real-world material surface morphology with holes into the generator network within the GAN. The model achieves high-precision contour reconstruction because, during the training phase, a discriminator network is introduced for adversarial gameplay. The discriminator continuously attempts to distinguish between the real, complete feed point cloud and the point cloud completed by the generator. To deceive the discriminator, the generator network is forced to deeply learn the physical angle of repose constraints, surface roughness distribution, and latent geometric manifold features of the boundary continuity of bulk feed stacks. Therefore, when a discrete morphology is input, the generator's encoder can extract local features of the boundary and global contextual information of the overall stack. The decoder, based on this internalized physical prior knowledge, directly samples and generates 3D data points that conform to physical laws in the latent space of the missing regions. The final output point cloud not only fills in the holes but also ensures a perfect fusion of the newly generated region with the original boundary in terms of geometric curvature, resulting in a corrected target spatial morphology.

[0126] S205. The target spatial shape is used as the real-time spatial shape of the current cycle to perform the step of calculating the incremental discharge volume of the current cycle based on the change of the spatial boundary between the real-time spatial shape of the current cycle and the real-time spatial shape of the previous cycle.

[0127] Spatial boundary change refers to the geometric decrease or volume reduction of the feed surface inside the truck compartment in terms of vertical height or three-dimensional spatial contour due to the unloading of feed between two adjacent time periods. This includes, but is not limited to, the Boolean difference volume of a height difference matrix or a three-dimensional mesh. Incremental discharge volume refers to the actual physical volume of feed unloaded from the bulk feed truck during the extremely short time interval between the previous period and the current period.

[0128] The system can calculate the incremental discharge volume using a grid-based differential integration method based on a digital elevation model (DEM). First, the system establishes a unified horizontal two-dimensional reference plane. The real-time spatial morphology of the previous cycle and the target spatial morphology of the current cycle are vertically projected onto this reference plane and divided into a dense, regular grid with identical resolution. For each grid cell, the system extracts the elevation values ​​at the center point of the grid for both cycles. The system iterates through all grids, subtracting the current cycle's elevation value from the previous cycle's elevation value to obtain the local height decrease at that grid location. Then, the system multiplies this height decrease by the physical base area of ​​a single grid cell to obtain the minute discharge volume of that local region. Finally, the system globally sums the minute discharge volumes of all grid cells to accurately calculate the incremental discharge volume of the entire carriage in the current cycle.

[0129] The system can also employ a technique based on 3D Delaunay triangulation and tetrahedral volume integration for calculation. The system places the real-time spatial morphology data points from the previous cycle and the target spatial morphology data points from the current cycle in the same 3D coordinate system. Using the 3D Delaunay triangulation algorithm, the system constructs a solid mesh model composed of countless non-intersecting 3D tetrahedra between these two layers of surface data points. This solid mesh perfectly fills the spatial boundary variation region between the two surfaces. For each generated tetrahedron, the system extracts the 3D coordinates of its four vertices and calculates the independent volume of the tetrahedron using a vector hybrid product algorithm. Finally, the system iterates through and accumulates the volumes of all tetrahedra in the solid mesh model; the sum obtained is the incremental discharge volume for the current cycle.

[0130] In the above embodiments, by extracting the spatial dynamic change features of the real-time spatial morphology in a continuous time series, the suspended obstruction area caused by unloading dust can be effectively identified and removed. Since dust will produce visual distortion of non-solid surfaces in the visual sensor, directly using morphological data containing distortion will lead to large errors in volume calculation. After removing the suspended obstruction area, the discrete real material surface morphology is obtained, and the spatial contour of the missing area is reconstructed based on the boundary continuity of the real material surface morphology. This can restore the real physical contour of the area affected by dust to the greatest extent, reduce the interference of dust on the visual detection results under harsh unloading conditions, improve the integrity and authenticity of the target spatial morphology, and thus improve the robustness and accuracy of incremental discharge volume calculation.

[0131] In the above embodiments, a general approach was proposed to peel off the suspended occlusion area and reconstruct the spatial contour of the resulting missing area. However, if simple geometric interpolation reconstruction is performed solely based on the static boundary remaining in the current cycle, it is often difficult to accurately reflect the dynamic collapse and slippage characteristics of the feed during unloading, easily leading to distortion of the reconstructed contour. To make the reconstructed spatial contour more consistent with the actual physical motion laws of bulk materials and further improve the fidelity of morphological restoration, this application provides a more accurate evolutionary method based on kinematic extrapolation for the spatial contour reconstruction step. The following is combined with... Figure 3 This application describes yet another method for detecting the unloading flow rate of bulk feed trucks based on visual recognition:

[0132] Please see Figure 3 This is another flowchart illustrating a method for detecting the unloading flow rate of a bulk feed truck based on visual recognition, as described in this application.

[0133] S301. Extract the spatial displacement vector of the discrete real material surface morphology between the current cycle and the previous cycle, and construct a bulk motion flow field that characterizes the slippage trend of the feed surface.

[0134] Bulk motion flow field refers to the vector set describing the velocity and direction of various local micro-elements in space during the unloading process of bulk materials. Its manifestations include, but are not limited to, three-dimensional dense optical flow field, point cloud scene flow, or discrete element velocity field. Feed surface slip tendency refers to the overall or local directional flow tendency of materials under the guidance of gravity or mechanical structures.

[0135] The system can implement this step using a 3D point cloud registration technique based on an iterative nearest-point algorithm. First, the system extracts discrete point cloud data of the real material surface morphology from the previous and current cycles, and uses a feature extraction algorithm to filter out key points with significant geometric features. Next, the system establishes an initial correspondence between the key points of the two cycles and solves for the global rigid body transformation matrix between the two frames of point clouds by iteratively minimizing the Euclidean distance between the corresponding points. Based on this, the system further employs a non-rigid body registration algorithm to calculate the precise 3D translation vector of each local data point from the previous cycle to the current cycle. These translation vectors are then interpolated and smoothed in space to construct a granular motion flow field covering the entire real material surface.

[0136] The system can also utilize a deep learning-based 3D scene flow estimation network to achieve this step. The system converts the real material surface morphology of two adjacent cycles into a voxel mesh or point cloud sequence, which is then input into the pre-trained scene flow estimation network. This model is able to extract displacement vectors and construct the flow field because it learns from a large amount of simulation data of bulk material flow during the training phase. The hierarchical feature extraction module and correlation calculation module within the model can deeply mine the deformation patterns of local geometric structures between adjacent frames. Through end-to-end computation, the network directly outputs the motion vector of each spatial point in the 3D coordinate system. By aggregating these output motion vectors, the system can construct a bulk material flow field characterizing the slippage trend of the feed surface.

[0137] S302. Extrapolate the kinematics of the bulk motion flow field from the boundary of the actual material surface shape to the interior of the missing region to predict the internal deformation field of the missing region in the current cycle.

[0138] The system extrapolates the flow field of bulk materials from the boundary of the actual material surface shape to the interior of the missing region kinematically, predicting the internal deformation field of the missing region in the current cycle. Specifically, this includes: extracting the edge displacement vector at the intersection of the actual material surface shape and the missing region as the boundary condition for flow field extrapolation; constructing a Poisson equation constrained by the edge displacement vector based on the assumptions of mass conservation and flow continuity of bulk materials; solving the Poisson equation to obtain the predicted displacement vector of each spatial node inside the missing region, and using all the predicted displacement vectors to constitute the internal deformation field of the missing region in the current cycle.

[0139] Kinematic extrapolation refers to the process of using the known motion state of a region, combined with physical laws or mathematical models, to infer the motion state of an unknown region. Internal deformation field refers to the set of vector distributions showing the positional changes of various spatial points within a missing region relative to the previous period in the current cycle. Edge displacement vector refers to the motion vector of valid data points immediately adjacent to the boundary of a data hole. Boundary conditions refer to the known variable requirements that must be satisfied at the boundary of the solution domain when solving differential equations. Poisson's equation is a partial differential equation commonly used to describe the distribution of potential fields; here, it is used to describe the smooth diffusion of the flow field in space. The assumptions of mass conservation and flow continuity state that within extremely short time periods, bulk materials will not be created or disappear out of thin air during flow, and their velocity field is continuous and without jumps in space.

[0140] The system can employ a numerical solution technique based on the finite difference method to implement this refinement scheme. First, the system discretizes the three-dimensional space containing the missing region and its boundaries into a regular 3D mesh. The system extracts the edge displacement vectors at the intersection of the actual material surface morphology and the missing region, assigning them to the corresponding boundary mesh nodes as Dirichlet boundary conditions for flow field extrapolation. In the Poisson equation constructed based on the assumptions of mass conservation and flow continuity, the system uses a central difference scheme to transform the Laplace operator in the equation into a discrete system of algebraic equations. Subsequently, the system solves this sparse linear system of equations using the conjugate gradient method or the Gauss-Seidel iteration method. After multiple iterations and convergence, the predicted displacement vectors of all mesh nodes within the missing region can be calculated. These vectors collectively constitute the internal deformation field for the current cycle.

[0141] The system can also employ finite element analysis (FEM) to achieve this refinement technique. The system meshes the missing region with three-dimensional Delaunay tetrahedral meshes, constructing an unstructured computational domain. The extracted edge displacement vectors are applied as boundary constraints to the surface nodes of the computational domain. For the constructed Poisson equation, the system derives its weak form using the Galerkin method and selects appropriate shape functions for each tetrahedral element, thereby assembling the global stiffness matrix and load vector. Finally, the system calls the direct solver to solve the global matrix equations, obtaining the predicted displacement vector for each node within the missing region. The predicted displacement vectors of all nodes accurately constitute the internal deformation field of the missing region in the current period.

[0142] S303. Apply the internal deformation field to the historical spatial morphology of the corresponding region in the previous cycle to generate the physical deformation prior contour of the missing region.

[0143] Historical spatial morphology refers to the three-dimensional surface data acquired and saved in the previous period, corresponding to the spatial location of the currently missing area. Physical deformation prior profile refers to a preliminary three-dimensional surface morphology generated by superimposing the predicted motion field onto historical data before final spatial profile reconstruction. It reflects the ideal motion result of materials without external interference, including but not limited to the deformed point cloud set or the evolved three-dimensional mesh. Evolution refers to the process by which data updates its state over time or according to specific rules.

[0144] The system can implement this step using a point cloud direct evolution technique based on 3D vector addition. First, the system reads historical spatial morphology point cloud data from memory, corresponding to the current missing region in the previous cycle. For each 3D coordinate point in this historical point cloud, the system uses nearest neighbor search or trilinear interpolation algorithms in the internal deformation field to query and obtain the predicted displacement vector corresponding to that point's location. Next, the system performs vector addition on the point's original 3D coordinates and the obtained predicted displacement vector to calculate the point's new spatial coordinates in the current cycle. After traversing all points in the historical spatial morphology and updating their coordinates, the system reassembles these updated coordinate points to evolve and generate the physical deformation prior contour of the missing region.

[0145] The system can also employ an implicit surface advection evolution technique based on the level set method to achieve this step. The system transforms the historical spatial morphology of the previous cycle into a symbolic distance field on a 3D mesh, serving as the initial implicit surface. The system uses the internal deformation field as the velocity field and constructs advection partial differential equations describing the surface's motion with the velocity field. The advection equations are numerically solved using an upwind difference scheme, pushing the symbolic distance field forward by one cycle's time step. After evolution, the system uses the moving cube algorithm to extract the zero isosurface from the updated symbolic distance field; this extracted new 3D surface is the physical deformation prior contour of the evolved missing region.

[0146] S304. Using the physical deformation prior contour as the reference shape, perform the step of reconstructing the spatial contour of the missing area caused by peeling based on the boundary continuity of the real material surface shape.

[0147] The baseline morphology refers to a three-dimensional model that serves as a reference standard or initial guess when performing data fitting, interpolation, or reconstruction. From the inventor's perspective, although the prior physical deformation profile generated through kinematic evolution is very close to the actual feed surface, due to the inevitable small errors in the prediction model, the edges of this prior profile may not perfectly align with the actual feed surface boundary observed in the current cycle. To eliminate such geometric discontinuities at the boundary and ensure seamless connection of the final morphology, the system uses this prior profile as the base for reconstruction. During the reconstruction process, the system no longer fills in gaps out of thin air, but uses the boundary data of the actual feed surface as a hard constraint to locally fine-tune and integrate the prior profile. This preserves the actual collapse morphology within the prior profile that conforms to the laws of granular mechanics, while ensuring the smooth continuity of the reconstructed area with the surrounding actual feed surface in terms of height and slope, ultimately outputting a target spatial morphology that combines physical realism and geometric integrity.

[0148] The system can implement this step using a Laplacian surface deformation algorithm. First, the system places the physical deformation prior contour and the discrete real material surface shape in the same coordinate system, identifies the overlapping nodes between the edge of the prior contour and the boundary of the real material surface, and calculates the spatial positional deviation between these nodes. The system uses the boundary nodes of the real material surface as fixed constraint points to construct a system of linear equations based on Laplacian coordinates. By solving this system of equations, the system smoothly diffuses the boundary positional deviation as an energy term to the internal nodes of the prior contour. During deformation, the algorithm can preserve the original local differential geometric features of the prior contour to the greatest extent possible, ensuring that the adjusted prior contour not only perfectly stitches its edges with the real material surface, but also maintains the evolved physical deformation features internally, thus achieving high-quality spatial contour reconstruction.

[0149] The system can also employ an error compensation technique based on radial basis function interpolation to achieve this step. The system extracts a series of sampling points at the intersection of the physical deformation prior contour edge and the actual material surface boundary, calculating the three-dimensional error vector for each sampling point between the two shapes. Using these sampling points as centers, the system constructs a radial basis function network, solves for the network weights using the error vector, and thus fits a continuous three-dimensional spatial error compensation field. Subsequently, the system applies this error compensation field to the entire physical deformation prior contour, correcting the coordinates of each point within the prior contour. Because radial basis functions decay with distance, the correction amount is maximized at the boundary to achieve seamless connection, while the correction amount decreases closer to the center of the missing region, thereby preserving the physical evolution results of the baseline shape to the greatest extent and completing the spatial contour reconstruction.

[0150] In the above embodiments, a flow field for bulk material motion is constructed by extracting the spatial displacement vectors of the actual material surface morphology between adjacent cycles. This flow field is then kinematically extrapolated from the boundary into the missing region to predict the internal deformation field. The internal deformation field is then applied to the historical spatial morphology of the corresponding region in the previous cycle, evolving to generate a priori physical deformation profile as a reconstruction benchmark. The flow of feed during unloading exhibits physical continuity and kinetic inertia. Extrapolating the displacement vectors of the actual material surface into the missing region can reasonably simulate the actual motion trend of the bulk material within the occluded area. Combining the predicted motion trend with the historical morphology to generate the priori profile ensures that the morphological reconstruction of the missing region is no longer limited to the static boundary geometric interpolation of the current cycle, but incorporates time-series evolution information consistent with the dynamics of bulk materials. This mechanism, combining historical morphology with kinematic prediction, makes the reconstructed spatial profile more closely match the actual physical deformation process of the feed surface, thereby improving the fidelity and reliability of the spatial profile reconstruction of the missing region.

[0151] The system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference needed]. Figure 4This is a schematic diagram of the physical device structure of a bulk feed truck unloading flow detection system based on visual recognition, provided in an embodiment of this application.

[0152] It should be noted that, Figure 4 The structure of the system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0153] like Figure 4 As shown, the system includes a Central Processing Unit (CPU) 401, which can perform various appropriate actions and processes based on a program stored in Read-Only Memory (ROM) 402 or a program loaded from storage portion 408 into Random Access Memory (RAM) 403, such as executing the methods described in the above embodiments. The RAM 403 also stores various programs and data required for system operation. The CPU 401, ROM 402, and RAM 403 are interconnected via a bus 404. An Input / Output (I / O) interface 405 is also connected to the bus 404.

[0154] The following components are connected to I / O interface 405: input section 406 including a camera, infrared sensor, etc.; output section 407 including a liquid crystal display (LCD) and speakers, etc.; storage section 408 including a hard disk, etc.; and communication section 409 including a network interface card such as a LAN (Local Area Network) card and a modem, etc. Communication section 409 performs communication processing via a network such as the Internet. Drive 410 is also connected to I / O interface 405 as needed. Removable media 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 410 as needed so that computer programs read from it can be installed into storage section 408 as needed.

[0155] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit (CPU) 401, it performs the various functions defined in the present invention.

[0156] It should be noted that the computer-readable medium shown in the embodiments of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. The transmitted data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof.

[0157] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0158] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the system described in the above embodiments; or it may exist independently and not assembled into the system. The storage medium carries one or more computer programs that, when executed by a processor of a system, cause the system to implement the methods provided in the above embodiments.

[0159] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0160] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0161] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0162] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for detecting the unloading flow rate of bulk feed trucks based on visual recognition, characterized in that, include: The baseline density is calculated based on the initial total mass and initial total volume of the bulk feed truck when it is fully loaded; Before the unloading mechanism is started, the initial spatial shape of the feed surface inside the bulk feed truck compartment is obtained; During the operation of the unloading mechanism, the real-time spatial morphology of the feed surface is acquired at preset time intervals; The incremental discharge volume for the current cycle is calculated based on the change in the spatial boundary between the real-time spatial morphology of the current cycle and the real-time spatial morphology of the previous cycle. The current cumulative discharge volume is calculated based on the spatial boundary difference between the real-time spatial morphology of the current cycle and the initial spatial morphology. Divide the current cumulative discharge volume by the initial total volume to obtain the unloading space evolution ratio; According to a preset mapping relationship, the evolution ratio of the unloading space is converted into a density compensation coefficient, and the density compensation coefficient is positively correlated with the evolution ratio of the unloading space. Multiply the reference density by the density compensation coefficient to obtain the instantaneous equivalent density for the current period; Multiply the incremental discharge volume by the instantaneous equivalent density to obtain the incremental unloading mass for the current cycle; The incremental unloading mass is accumulated to obtain the real-time cumulative unloading total mass.

2. The method according to claim 1, characterized in that, The step of converting the unloading space evolution ratio into a density compensation coefficient according to a preset mapping relationship specifically includes: The evolution ratio of the unloading space is mapped to the relative depth feature of the current unloading interface; Based on the preset distribution law of bulk lateral pressure, the local compressive stress corresponding to the relative depth characteristic is calculated. The local compressive stress is used to characterize the compaction effect of the feed weight above on the current depth. According to the preset stress density conversion model, the local extrusion stress is converted into the compaction deformation rate; The density compensation coefficient is obtained by correcting the preset benchmark compensation coefficient based on the compaction deformation rate.

3. The method according to claim 1, characterized in that, The step of converting the unloading space evolution ratio into a density compensation coefficient according to a preset mapping relationship specifically includes: Extract the surface geometric features of the real-time spatial morphology of the current period; The spatial dispersion of the surface geometric features is calculated to obtain the morphological undulation index, which is used to characterize the degree of ruggedness of the feed surface caused by local flow during the unloading process. The dynamic loosening coefficient of the current unloading interface is determined based on the difference between the morphological undulation index and the preset flatness benchmark value. The density compensation coefficient is obtained by combining the preset basic compensation value corresponding to the evolution ratio of the unloading space with the dynamic loosening coefficient.

4. The method according to claim 1, characterized in that, After acquiring the real-time spatial morphology of the feed surface at preset time intervals during the operation of the unloading mechanism, the method further includes: Extract the spatial dynamic change characteristics of the real-time spatial morphology in a continuous time series; Based on the spatial dynamic change characteristics, the suspended occlusion area in the real-time spatial form is identified, and the suspended occlusion area represents the visual distortion of non-physical surfaces caused by unloading dust. By stripping away the suspended occlusion area in the real-time spatial morphology, a discrete real material surface morphology is obtained. Based on the boundary continuity of the actual material surface morphology, the spatial contour of the missing area caused by peeling is reconstructed to obtain the corrected target spatial morphology. The target spatial shape is used as the real-time spatial shape of the current cycle to perform the step of calculating the incremental discharge volume of the current cycle based on the spatial boundary change between the real-time spatial shape of the current cycle and the real-time spatial shape of the previous cycle.

5. The method according to claim 4, characterized in that, The step of identifying the suspended occlusion region in the real-time spatial form based on the spatial dynamic change characteristics specifically includes: From the aforementioned spatial dynamic change characteristics, the depth jump frequency and spatial motion vector of each local region in the real-time spatial morphology are analyzed; Local regions with depth jump frequencies greater than a preset frequency threshold are selected as candidate occlusion regions; Calculate the corresponding directional dispersion based on the spatial motion vector within the candidate occlusion region; Candidate occlusion regions with directional dispersion greater than a preset dispersion threshold are identified as floating occlusion regions.

6. The method according to claim 4, characterized in that, Before reconstructing the spatial contour of the missing area caused by peeling based on the boundary continuity of the actual material surface morphology to obtain the corrected target spatial morphology, the method further includes: Extract the spatial displacement vector of the discrete real material surface morphology between the current cycle and the previous cycle to construct a bulk motion flow field characterizing the slippage trend of the feed surface; The kinematic extrapolation of the granular motion flow field from the boundary of the actual material surface shape into the interior of the missing region is performed to predict the internal deformation field of the missing region in the current cycle. The internal deformation field is applied to the historical spatial morphology of the corresponding region in the previous cycle to generate the physical deformation prior contour of the missing region. Using the physical deformation prior contour as the reference shape, the step of reconstructing the spatial contour of the missing area generated by peeling is performed based on the boundary continuity of the actual material surface shape.

7. The method according to claim 6, characterized in that, The step of kinematically extrapolating the flow field of the loose material from the boundary of the actual material surface shape into the interior of the missing region to predict the internal deformation field of the missing region in the current cycle specifically includes: Extract the edge displacement vector at the boundary between the actual material surface shape and the missing region, and use it as the boundary condition for flow field extrapolation; Based on the assumptions of mass conservation and flow continuity of bulk materials, a Poisson equation constrained by the aforementioned edge displacement vector is constructed. Solving the Poisson equation yields the predicted displacement vectors of each spatial node within the missing region, and all the predicted displacement vectors constitute the internal deformation field of the missing region in the current period.

8. A bulk feed truck unloading flow detection system based on vision recognition, characterized in that, The system includes: One or more processors and a memory; the memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the system, the system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the system, the system performs the method as described in any one of claims 1-7.

Images