Method and device for accurately controlling multi-variety dynamic sample separation quantity

Abstract

According to the multi-variety dynamic sample separation amount accurate control method and device provided by the embodiment of the invention, the multi-feature fusion model is innovatively designed, and accurate identification of material varieties is realized through shape analysis and spectrum detection. And a discharging control system is constructed, and a reliable weight monitoring mechanism is established in combination with multiple compensation. Adaptive learning is introduced, and the accuracy of sample separation is ensured through parameter updating and predictive turn-off. According to the method, the defects of the traditional technology in the aspects of material identification, blanking control, turn-off precision and the like are effectively overcome, and a technical guarantee is provided for dynamic sample separation.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a method and apparatus for precise control of dynamic sample quantity for multiple varieties. Background Technology

[0002] Existing dynamic sample quantity control methods have significant shortcomings. Traditional systems perform poorly in material identification and parameter control, failing to effectively integrate image features and spectral data, thus affecting the accuracy of sample dispensing.

[0003] Furthermore, existing technologies suffer from bottlenecks in material feeding control and weight monitoring. Most systems lack sophisticated volume calculation mechanisms and compensation strategies, leading to inaccurate sample quantity control.

[0004] The existing system has technical shortcomings in shutdown control. The lack of in-depth analysis of system latency makes it difficult to achieve accurate material shut-off through predictive mechanisms, affecting the reliability of sample separation. Solving these problems is crucial for improving sample separation efficiency. Summary of the Invention

[0005] To address the problems in the existing technology, this application provides a method and device for precise control of dynamic sampling quantity for multiple varieties, which can effectively solve the shortcomings of traditional technologies in material identification, feeding control and shut-off accuracy, and provide technical support for dynamic sampling.

[0006] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides a method for precise control of dynamic sampling quantity for multiple varieties, including: Grain images and near-infrared spectral data are collected. Edge extraction and shape analysis are performed on the image data to extract the area, perimeter, and roundness features of the grain particles. Baseline correction and standardization are performed on the near-infrared spectral data. The shape features and spectral features are input into the variety identification model to identify the current material variety. Based on the material variety, the corresponding bulk density parameters, filling coefficient parameters, and shape influence coefficient parameters are retrieved from the parameter database. The parameters and the target feeding weight are input into the feeding control model. The target discharge volume is calculated based on the bulk density parameter. The target discharge volume is divided by the discharge volume per single turn of the rotary valve to obtain the integer number of turns. The remaining volume is calculated. The final opening angle is calculated based on the filling coefficient parameter and the shape influence coefficient parameter. The opening angle and the rotation speed parameter are input into the terminal control model to calculate the opening time. At the same time, multi-layer material level sensor data is collected. The current discharge weight is calculated based on the material level data and the bulk density parameter. The discharge weight is monitored in real time. The difference between the current feeding weight and the target feeding weight is input into the compensation model to calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. The compensation weight is subtracted from the target feeding weight to obtain the shut-off threshold. The weight growth rate is calculated, and the time when the shut-off threshold is reached is predicted based on the weight growth rate. The electromagnetic braking mechanism is activated in advance to shut off the feeding port, and the material is introduced into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, the deviation between the actual feeding weight and the target weight is transmitted to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

[0007] Furthermore, it also includes: acquiring high-resolution images of grain samples under standard light source illumination, performing Gaussian filtering on the images to eliminate noise, using the Canny operator to perform edge detection to extract particle contours, converting the contours into binary images, performing morphological operations on the binary images to remove edge burrs, performing particle segmentation based on connected component analysis, and calculating the pixel area, boundary perimeter, and roundness index of each particle region after segmentation as shape features. The light source of the near-infrared spectral acquisition device is turned on to scan the grain sample and obtain reflectance spectral data. The dark current signal is subtracted from the reflectance spectral data, whiteboard calibration is performed to eliminate instrument drift, and baseline offset is removed by multivariate scattering correction method. The corrected spectral data is resampled according to a preset wavelength interval, and the resampled data is subjected to standard normal transformation to generate standardized spectral features.

[0008] Furthermore, it also includes: concatenating the shape features of the grain sample with the standardized spectral features to construct a feature vector, performing principal component analysis to reduce the dimensionality of the feature vector, inputting the dimensionality-reduced features into a pre-trained convolutional neural network to extract local feature maps, inputting the feature maps into a fully connected layer to calculate the probability score of each variety, and determining the material variety identifier based on the highest probability score; Based on the material type identifier, the corresponding record is retrieved from the parameter database, the basic bulk density value corresponding to the type is read, the basic bulk density is multiplied by the temperature and humidity correction coefficient to obtain the real-time bulk density parameter, the valve filling coefficient value and shape influence coefficient value corresponding to the type are read, and the real-time bulk density parameter, filling coefficient value and shape influence coefficient value are written into the parameter cache area of ​​the control model.

[0009] Furthermore, it also includes: calculating the target feeding volume by the ratio of the target feeding weight to the real-time bulk density parameter; dividing the target feeding volume by the product of the number of valve cells and the volume of a single cell to obtain the theoretical feeding volume of a single valve turn; calculating the actual feeding volume of a single valve turn based on the product of the theoretical feeding volume of a single valve turn and the filling coefficient parameter and the shape influence coefficient parameter; and dividing the target feeding volume by the actual feeding volume of a single valve turn and rounding down to obtain the integer number of turns. The remaining volume is obtained by subtracting the product of the integer number of turns and the actual discharge volume per turn of the valve from the target discharge volume. The remaining volume is then divided by the volume of a single cell to calculate the number of cells to be opened. The number of cells to be opened is multiplied by the angle corresponding to a single cell to obtain the theoretical opening angle. Finally, the theoretical opening angle is calculated by multiplying it by the shape influence coefficient parameter.

[0010] Furthermore, it also includes: calculating the basic rotation time by dividing the final opening angle by the rotation speed of the rotary valve; taking the sum of the basic rotation time and the material falling time as the theoretical opening time; adding the sum of the sensor response delay time and the control system delay time to the theoretical opening time to obtain the actual opening time; and setting the operating parameters of the servo motor and the start-up sequence of the electromagnetic braking mechanism according to the actual opening time. Read the material level height data collected by the multi-layer material level sensor array, calculate the distance between each sensor position and the current material level height, assign sensor weight coefficients according to the distance, sum the product of the material level height data and the weight coefficients to obtain the weighted average material level height, and calculate the current material discharge weight by multiplying the weighted average material level height by the bottom area of ​​the receiving device, the real-time bulk density parameter, and the material surface distribution correction coefficient.

[0011] Furthermore, it also includes: subtracting the target material weight from the current material weight to obtain a weight difference; setting a compensation coefficient based on the weight difference; multiplying the product of the valve opening area and the drop height by the real-time bulk density parameter to obtain the air material compensation weight; multiplying the product of the material flow rate and the system response time by the real-time bulk density parameter to obtain the system delay compensation weight; multiplying the brake overshoot volume by the real-time bulk density parameter to obtain the brake overshoot compensation weight; and adding the air material compensation weight, the system delay compensation weight, and the brake overshoot compensation weight to obtain the total compensation weight. Collect the material feeding weight data of two adjacent sampling periods, calculate the difference between the material feeding weight data and divide it by the sampling time interval to obtain the weight growth rate, subtract the total compensation weight from the target material feeding weight to obtain the shutdown threshold, multiply the shutdown threshold by the system safety factor and subtract the minimum material feeding weight threshold to obtain the actual shutdown threshold, and calculate the time required for the current material feeding weight to reach the actual shutdown threshold based on the weight growth rate.

[0012] Furthermore, it also includes: multiplying the weight growth rate by the time required to reach the shut-off threshold and adding the current feeding weight to obtain the predicted weight; calculating the early shut-off time based on the predicted weight; subtracting the electromagnetic braking time, the control system delay time, and the material falling time from the early shut-off time to obtain the actual shut-off trigger time; controlling the servo motor to stop and starting the electromagnetic brake to shut off the feeding port at the actual shut-off trigger time; and driving the feeding distributor to transport the feeding material to the bulk density measuring container, the moisture measuring container, and the imperfect particle measuring container respectively. The weight deviation is obtained by subtracting the target weight from the actual weight of the material. The relative error is obtained by calculating the ratio of the weight deviation to the target weight. The filling coefficient correction and shape influence coefficient correction are calculated based on the relative error. The filling coefficient correction is subtracted from the original filling coefficient parameter to obtain the updated filling coefficient parameter. The shape influence coefficient correction is subtracted from the original shape influence coefficient parameter to obtain the updated shape influence coefficient parameter.

[0013] Secondly, this application provides a device for precise control of dynamic sample dispensing volume for multiple varieties, comprising: The feeding model determination module is used to collect grain images and near-infrared spectral data, perform edge extraction and shape analysis on the image data, extract the area, perimeter, and roundness features of the grain particles, perform baseline correction and standardization processing on the near-infrared spectral data, input the shape features and spectral features into the variety identification model, identify the current material variety, and retrieve the corresponding bulk density parameters, filling coefficient parameters, and shape influence coefficient parameters from the parameter database according to the material variety, and input the parameters and target feeding weight into the feeding control model; The feeding operation control module is used to calculate the target feeding volume based on the bulk density parameter, divide the target feeding volume by the feeding volume per turn of the rotary valve to obtain the integer number of turns, calculate the remaining volume, calculate the final opening angle according to the filling coefficient parameter and the shape influence coefficient parameter, input the opening angle and the rotation speed parameter into the terminal control model to calculate the opening time, and simultaneously collect data from multi-layer material level sensors, calculate the current feeding weight according to the material level data and the bulk density parameter, and monitor the feeding weight in real time. The adaptive shutdown module is used to input the difference between the current feeding weight and the target feeding weight into the compensation model, calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. The shutdown threshold is obtained by subtracting the compensation weight from the target feeding weight. The weight growth rate is calculated, and the time when the shutdown threshold is reached is predicted based on the weight growth rate. The electromagnetic braking mechanism is activated in advance to shut off the feeding port, and the material is introduced into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, the deviation between the actual feeding weight and the target weight is transmitted to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

[0014] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the multi-variety dynamic sample quantity precise control method.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for precise control of dynamic sample quantity for multiple varieties.

[0016] Fifthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implements the steps of the multi-variety dynamic sample quantity precise control method.

[0017] As can be seen from the above technical solution, this application provides a method and device for precise control of dynamic sampling quantity for multiple varieties. Through innovative design of a multi-feature fusion model, and by using shape analysis and spectral detection, accurate identification of material varieties is achieved. A feeding control system is constructed, and a reliable weight monitoring mechanism is established by combining multiple compensations. Adaptive learning is introduced, and the accuracy of sampling is ensured through parameter updates and predictive shutdown. This method effectively solves the shortcomings of traditional technologies in material identification, feeding control, and shutdown accuracy, providing technical support for dynamic sampling. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the method for precise control of dynamic sampling quantity for multiple varieties in this application embodiment; Figure 2 This is a structural diagram of the multi-variety dynamic sampling quantity precision control device in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0020] Figure label: Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0023] To address the problems existing in current technologies, this application provides a method and apparatus for precise control of dynamic sampling quantities for multiple product types. Through an innovative design of a multi-feature fusion model, and by utilizing shape analysis and spectral detection, accurate identification of material types is achieved. A feeding control system is constructed, incorporating multiple compensations to establish a reliable weight monitoring mechanism. Adaptive learning is introduced, and through parameter updates and predictive shutdown, the accuracy of sampling is ensured. This method effectively solves the shortcomings of traditional technologies in material identification, feeding control, and shutdown accuracy, providing technical support for dynamic sampling.

[0024] To effectively address the shortcomings of traditional technologies in material identification, feeding control, and shut-off accuracy, and to provide technical support for dynamic sampling, this application provides an embodiment of a method for precise control of dynamic sampling quantities for multiple product varieties. See [link to embodiment]. Figure 1 The method for precise control of dynamic sampling quantity for multiple varieties specifically includes the following: Step S101: Collect grain images and near-infrared spectral data, perform edge extraction and shape analysis on the image data, extract the area, perimeter, and roundness features of the grain particles, perform baseline correction and standardization on the near-infrared spectral data, input the shape features and spectral features into the variety identification model, identify the current material variety, retrieve the corresponding bulk density parameters, filling coefficient parameters, and shape influence coefficient parameters from the parameter database according to the material variety, and input the parameters and target feeding weight into the feeding control model; Optionally, this embodiment focuses on the online sampling equipment of the grain depot receiving line in step S101. In this embodiment, a standard light source box and an industrial camera are combined above the feeding chute, with the field of view covering a single-layer thin-layer instantaneous material strip.

[0025] To ensure reliable shape measurements, white balance was achieved using a gray card before shooting. Exposure was initially determined using automatic metering and then locked to prevent color drift between batches from affecting spectral channel mapping. The original image underwent a light Gaussian filter to suppress random noise without altering edges. Canny was then used to extract grain contours, and the edge output was intersected with a coarse segmentation mask generated by Otsu thresholding to subtract background reflections. Connected component analysis identified contacting grains, and morphological opening operations were used to segment adhesion points. Skeleton and concave point detection served as dual criteria to avoid over-segmentation. The pixel area and boundary length of each connected component were calculated, and roundness was determined using 4πA / P. 2 This dimensionless index, where A represents the projected area of ​​the particle and P is the corresponding perimeter, naturally reflects the degree to which the particle shape deviates from a circle. The value will be lower for slender or broken particles, which is closer to the intuition of "morphological differences affecting filling" in the sampling system in this embodiment.

[0026] This embodiment does not rely solely on vision as the sole criterion. An integrated near-infrared probe on the chute sidewall acquires the reflectance spectrum through a progressive scanning motion as material passes through. The acquired raw spectrum is first darkened by a current filter, then calibrated using a whiteboard to offset light source drift. Next, MSC or SNV is applied to remove baseline fluctuations caused by scattering. The bands are resampled with a preset step size, and saturated peaks are removed. In this embodiment, the local energy of protein and starch-related absorption peaks is preserved for a simple reason: different varieties exhibit stable differences in these chemical compositions, meaning that even with similar appearances, the spectra will provide independent chains of evidence. Visual shape features and standardized spectral features are aligned across the sample dimension to form a spliced ​​vector that is then incorporated into the variety identification model.

[0027] The variety identification model in this embodiment adopts a dual-pathway structure of "shape tower + spectral tower".

[0028] The shape tower receives statistical values ​​(mean, variance, quantiles) of area, perimeter, and roundness distribution, along with a major axis estimate and the perimeter normalized to area. These are then compressed into a low-dimensional shape embedding using two fully connected layers and ReLU. The spectral tower extracts absorption patterns within local bandwidth using a one-dimensional convolutional network, followed by global average pooling to obtain the spectral embedding. The two embeddings are concatenated and fed into an attention-based fusion layer. The attention weights are generated by the batch image quality score and the spectral signal-to-noise ratio. This ensures that the model automatically prioritizes the spectrum when there is significant dust or poor image edges, and vice versa. The output is a set of variety probabilities; the variety identifier corresponding to the highest probability is selected.

[0029] In this embodiment, the reason for emphasizing dual-source fusion is that the field conditions are complex: late-season rice grains are small and slender with low roundness; in some places, wheat and black wheat are similar in shape but still have spectral differences. Including both paths will prevent being misled by a single path.

[0030] In this embodiment, upon obtaining the variety identifier, the parameter database is immediately accessed to retrieve the variety's basic bulk density, rotary valve filling factor, and shape influence factor, which are then combined with current temperature and humidity sensor data to generate real-time parameters. The bulk density varies slightly with moisture content and temperature. In this embodiment, the basic bulk density is multiplied by an environmental correction factor to obtain the real-time bulk density used for volume conversion. The filling factor reflects the actual filling rate within the rotary valve cell. The shape influence factor, derived from upstream shape statistics, is equivalent to a fine-tuning weight for the opening angle-discharge volume relationship. Long or irregularly shaped particles are more likely to form arches within the cell cavity, requiring a slightly larger opening angle to achieve the same volume. These parameters are loaded into the feed control model's cache; subsequent steps, such as volume decomposition, integer number of turns, and remaining cell calculations, all depend on these values.

[0031] This embodiment emphasizes the natural correlation between shape characteristics and subsequent control quantities because: reduced roundness usually means a longer shape or broken edges, resulting in poorer flowability and a lower filling rate. The shape influence coefficient should be increased accordingly to compensate for insufficient opening angle. Furthermore, once spectral characteristics provide a more stable variety determination, the base value of bulk density is no longer estimated but derived from statistical records of the variety in historical samples. This avoids ignoring density differences between different varieties and prevents incorrect volume conversion. For example, the difference in the proportion of amylopectin between japonica and indica rice is reflected in near-infrared absorption. In this embodiment, the spectrometer in the model is sensitive to this difference, so the database retrieves two different sets of density and filling parameters, naturally separating the subsequent volume-circle calculation path.

[0032] This embodiment also handles two types of abnormal situations.

[0033] Firstly, the proportion of varieties in the mixed grain flow changes abruptly, and both the image statistics and spectral distribution show double peaks in a short period of time. The highest probability output by the model is no longer stable. In this embodiment, the entropy threshold within the short window is introduced as a "variety uncertainty" indicator. The control model then delays parameter switching and uses a transition coefficient for gradual interpolation to avoid sudden changes in the opening angle caused by instantaneous switching.

[0034] Secondly, surface dust and powder coverage can increase the roundness of the image and the edge burrs. In this embodiment, a mass fraction screening based on the area ratio of concave points is placed in front of the shape tower to process particles with abnormally high concavity with low weight, so as to prevent dust and powder artifacts from being regarded as real shapes and affecting the shape influence coefficient.

[0035] In this embodiment, the results of the above identification and parameter retrieval are finally sent to the material feeding control model. The model's input includes the target material feeding weight and three types of real-time parameters. The output does not directly provide the opening time, but instead projects the quality target onto the target volume and then passes it to the volume decomposition process in S102. The connection here is crucial: if the preceding product type and parameters are misjudged, the subsequent integer number of loops and the remaining grid will deviate. In this embodiment, the deviation of "actual material feeding weight - target weight" is recorded on the control loop and fed back to the adaptive learning module to continuously correct the filling coefficient and shape influence coefficient. The direction follows physical common sense—a lighter weight indicates insufficient filling or poor shape flowability, requiring an increase in the shape influence coefficient; a heavier weight indicates the opposite.

[0036] This embodiment alleviates two practical pain points in terms of technical effectiveness.

[0037] Firstly, when multiple varieties are mixed, the traditional method of feeding materials with fixed parameters often amplifies the error to the downstream of the process. In this embodiment, visual + spectral recognition is used to first determine the variety, and then bring out the density and equipment side parameters of the variety, so that the source of error can be suppressed.

[0038] Secondly, the relationship between shape and filling is often handled empirically. In this embodiment, objective features such as roundness and circumference are used as inputs, so that the influence of shape is changed from "feeling" to "data calculation", which makes it more stable for changes in varieties or seasons.

[0039] For example, for early indica rice batches, the roundness distribution in the image is lower, and the absorption peak position in a specific band of the spectrum is regularly shifted relative to that of southern japonica rice. After fusion, it is identified as indica rice. The bulk density and filling coefficient returned by the database are slightly lower, and the shape influence coefficient is higher. Based on these parameters, the subsequent target volume and opening angle will be slightly enlarged compared to japonica rice. Another type is a shipment of black wheat mixed with ordinary wheat. The shape difference is not significant, but the absorption intensity in the protein-related band of near-infrared radiation changes significantly. The model can still stably give a "wheat-high protein" class, and the bulk density version retrieved from the parameter library is slightly higher, avoiding the overestimation of volume due to conversion with a lower density. The entire process is seamless, with a tight logical chain of identification-parameter-control, making it suitable for long-term operation on actual receiving lines.

[0040] Step S102: Calculate the target feeding volume based on the bulk density parameter, divide the target feeding volume by the feeding volume per single turn of the rotary valve to obtain the integer number of turns, calculate the remaining volume, calculate the final opening angle based on the filling coefficient parameter and the shape influence coefficient parameter, input the opening angle and the rotation speed parameter into the terminal control model to calculate the opening time, and simultaneously collect data from the multi-layer material level sensor, calculate the current feeding weight based on the material level data and the bulk density parameter, and monitor the feeding weight in real time; Optionally, this embodiment focuses on the sampling platform at the acquisition point in S102. The scenario is that the samples from the mixed-variety inbound truck need to be allocated to three types of testing containers for bulk density, moisture, and imperfect particles according to the target material weight.

[0041] In step S101, the current material type has been identified using image and near-infrared modeling, and the bulk density ρ, fill factor k_f, and shape influence factor k_s have been retrieved from the parameter library. The target feed weight W_t has also been assigned by task scheduling. The calculation unit first divides W_t by ρ to obtain the target feed volume V_t. The logic is simple: the same weight corresponds to different volumes for materials with different densities. If the valve action is not planned in the volume domain, the number of rotations and angles will be out of sync with the actual mass. The equipment file records the valve structure parameters: number of cells N_cell, nominal volume of a single cell V_cell, and theoretical volume of one rotation N_cell·V_cell.

[0042] Considering the differences in particle shape and flowability among the filling materials within the rotary valve compartments, the actual material volume per turn is calculated using the formula V_rev = N_cell·V_cell·k_f·k_s. Dividing V_t by V_rev and rounding down yields the integer number of turns n_rev. The remaining amount, V_rem = V_t − n_rev·V_rev, is then converted to the number of compartments to be opened, n_cell = V_rem / V_cell. This is then converted to the theoretical opening angle θ_th = n_cell·(360° / N_cell) based on the toothed disc angle scale. The shape influence coefficient k_s reflects the change in air gap rate caused by the aspect ratio and friction angle; therefore, the final opening angle θ = θ_th·k_s, where k_s > 1 indicates that the angle is appropriately increased for slender particles to compensate for "underfilling".

[0043] In this embodiment, the opening time is derived by inputting θ and the rotational speed parameter ω into the terminal control model. The basic rotation time of the rotary valve is t_rot = θ / ω. However, the material also has a falling time t_fall (determined by the free fall distance from the hopper opening to the receiving device and air resistance) and a response delay t_lat of the sensor and control link (encoder sampling, motor driver, and relay delay). The actual opening time is obtained by combining these factors: t_open = t_rot + t_fall + t_lat. t_fall is determined by the calibrated fall distance h and the equivalent damping coefficient table. In damp batches or dusty conditions, t_fall will be slightly longer. The parameter library provides interpolation based on humidity and dust rate. The reason for this setting is that the physical rotation of the valve does not mean that the weight immediately increases to the required level. If the time delay is ignored, a systematic deviation will be seen in the real-time monitoring stage, and the subsequent compensation model will be under too much pressure.

[0044] In this embodiment, while performing the opening action, data from a multi-layer material level sensor array installed on the wall of the receiving device is collected. The sensor layout is distributed from bottom to top in an equidistant or non-equidistant manner, with each sensor returning the distance d_i relative to the material interface. This distance is then geometrically converted to obtain the local material level h_i. Considering that the cone inside the container is not a horizontal plane, in this embodiment, each layer is assigned a weight w_i. The weight and the height of the sensor correspond to the cone model: readings near the top of the cone are more sensitive but also have a larger variance, so the weight is slightly lower; the middle region is more stable and has a higher weight. The weighted average material level h̄ = Σ(w_i·h_i) / Σw_i is then multiplied by the effective bottom area A of the receiving device, the material surface distribution correction coefficient γ, and the real-time density ρ to obtain the current feeding weight estimate W_c = ρ·A·h̄·γ. γ is mapped from the friction state between the cone's half-apex angle and the wall surface. When fine powder adheres to the wall surface, the cone shape is steeper, and γ is adjusted accordingly.

[0045] In this embodiment, W_c is placed in the real-time monitoring loop and recorded together with the encoder angle and timer to form a three-dimensional trajectory of "angle-time-weight".

[0046] In this embodiment, the monitoring algorithm performs a Kalman filter on W_c within a short time window. The observation is based on the material level to estimate the mass, and the process model is a mass growth rate derived from the known ω and ρ, offsetting sensor jitter and occasional splashes. If the growth rate of W_c deviates from the theoretical slope derived from θ and ω, the system first checks for instantaneous bridging or sudden changes in material flow lubrication: the appearance of bridging will prevent the upper material level from decreasing, but the mass at the bottom will not increase. The algorithm identifies this through the difference between the lower and upper sensors. Once confirmed, ω is briefly increased and the remaining angle of θ is finely adjusted to break the bridging and return to the original trajectory. This processing follows the natural flow pattern: particles easily form arches under certain powder content and wall friction, and short disturbances can resolve this.

[0047] This embodiment also performs online verification of the shape influence coefficient k_s at this stage. In this embodiment, the ratio of the actual short-window quality increment ΔW_c to the theoretical increment ΔW_th is used to check whether k_s is too small or too large. If the deviation persists and is not caused by sensor noise, a slight temporary correction to k_s is given for this task, but it is not immediately written back to the parameter library to avoid a single accidental operating condition changing the long-term parameters. Only after the task is completed is the actual deviation processed by the adaptive learning model of S107. The reason for the two-layer approach is that changes in the field operating conditions are generally instantaneous and reversible, and long-term parameter updates require more batches of evidence.

[0048] This embodiment takes into account the typical differences between different varieties and provides two practical scenarios for supplementary explanation. The first is long-grain indica rice, which has a slender shape and k_s is often greater than 1. If executed according to θ_th, underfilling will occur. In this embodiment, a smaller ω is maintained during the opening time, allowing more time for the particles to fill the cavity, resulting in a smoother W_c curve. The second is waxy corn, which has a short grain shape and high waxiness, resulting in poor flowability and potentially a lower k_f. For the same target volume, a larger opening angle is required. The level sensor is more sensitive to fluctuations in the upper layer, and increasing the process noise covariance of the filter is more prudent. The underlying physical relationship between the two examples is consistent: the particle shape and surface properties change the filling and flow rate, which is ultimately reflected in θ and t_open. The monitoring closed loop digests these differences in real time.

[0049] The technical advantages of this embodiment are reflected in the fact that all control quantities are expressed as a "density-volume-angle / time-weight" chain. The breakdown of the target volume (integer number of revolutions and remaining angle) balances action granularity and achievable accuracy within structural constraints; the weighted fusion of multiple material levels incorporates non-ideal stacking shapes within the container into the estimation, without relying on expensive dynamic load weighing; real-time trajectory comparison and bridge arch identification resolve flow anomalies as early as possible, reducing the burden of subsequent compensation. The final data stream (θ, ω, t_open, W_c curve, temporary k_s correction) provides clean and traceable input for downstream compensation and self-learning, maintaining a rigorous technical connection with preceding product identification and parameter retrieval.

[0050] Step S103: Input the difference between the current feeding weight and the target feeding weight into the compensation model, calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. Subtract the compensation weight from the target feeding weight to obtain the shut-off threshold. Calculate the weight growth rate. Predict the time when the shut-off threshold will be reached based on the weight growth rate. Activate the electromagnetic braking mechanism in advance to shut off the feeding port. Import the material into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, transmit the deviation between the actual feeding weight and the target weight to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

[0051] Optionally, in this embodiment, after the initial calculation of θ and t_open is given in S102, the closed-loop correction is performed in S103.

[0052] In this embodiment, the difference between the real-time estimated current feeding weight W_c and the target feeding weight W_t is used to obtain ΔW=W_t−W_c. This difference is not directly used to shut off the valve because there is a non-negligible physical tail between the valve closing and the material completely stopping.

[0053] In this embodiment, the compensation model is divided into three categories: airborne material, system delay, and braking overshoot. Airborne material compensation comes from particles in the free fall section below the valve orifice and above the container; system delay compensation comes from the response time of the sensing chain and the execution chain; braking overshoot compensation describes the volume corresponding to the remaining opening of the valve plate due to inertia and material rebound after the electromagnetic braking is triggered.

[0054] This embodiment employs a dual-channel approach—parameter-observation—to calculate aerial material compensation. The parameter channel analytically expresses the current valve opening area S, equivalent fall height h, and real-time bulk density ρ as inputs to estimate the aerial volume. V_air≈S·√(2h / g)·τ, Where τ is the average flow velocity correction time constant within the short window, and g is the gravitational acceleration; multiplying by ρ yields the airborne material compensation weight W_air. The observation channel originates from the upper-lower relative phase of the material level array. If the upper sensor descends first while the lower sensor has not yet changed, it indicates that there is a section of material flow "in transit". In this embodiment, this phase difference is used to make a small-range correction to τ to avoid misjudging occasional material tongues as stable flow. The system delay compensation W_lat is given by multiplying the material flow velocity v_flow by the total response time t_sys (including encoder sampling, PLC cycle, driver and relay delays) and then multiplying by ρ; the brake overshoot compensation W_ov is obtained by multiplying the brake overshoot volume V_ov (the open-loop calibration curve is mapped from θ, ω and motor inertia) by ρ. The three terms are added together: W_comp = W_air + W_lat + W_ov, and the turn-off threshold W_th = W_t − W_comp. W_air, W_lat, W_ov, and W_th respectively represent the weight in the air, the additional falling weight caused by system delay, the braking overshoot weight, and the target mass threshold that needs to be triggered in advance.

[0055] In this embodiment, to determine when to trigger shutdown, it is also necessary to know how fast the mass increases. In this embodiment, two adjacent W_c samples are read at a fixed sampling interval Δt, the difference is taken to obtain ΔW_c, and this difference is divided by Δt to obtain the weight increase rate r = dW / dt. r is not constant and will slowly drift with the progress θ and the wetness of the material flow. In this embodiment, r is exponentially weighted, giving higher weight to the most recent moment, so that it keeps up with changes and is not dragged down by noise. Based on this, the time to reach the shutdown threshold is predicted as t_hit = (W_th − W_c) / r. t_hit is not the final trigger time. In this embodiment, the combined lag of the electromagnetic brake coil actuation time t_b, the control system delay t_sys, and the fall time t_fall is also deducted to obtain the actual shutdown trigger time t_trig = t_hit − (t_b + t_sys + t_fall). The scheduler issues a stop and brake command at t_trig, the valve begins to close, the material flow ends near t_hit, and the weight in the container approaches W_t without overshooting.

[0056] In this embodiment, after shutdown, the distributor directs the remaining material to a predetermined container. The distribution rules for the three channels—bulk density, moisture, and imperfect particles—are preset by the task order. The valve positions and baffles operate in an interlocked manner to prevent mixed loading due to concurrent paths. In this embodiment, W_c continues to be sampled during braking, recording the wake curve of the short window after shutdown. If the wake is abnormally long, it is attributed to dust adhesion or wet batch stringing. The system increases the nominal value of W_ov in the next cycle to avoid repeated deviations. The logic here is consistent with natural phenomena: increased adhesion leads to excessive overshoot, and the parameters should be adjusted upwards appropriately.

[0057] This embodiment does not rely entirely on open-loop calibration for the source of compensation parameters. Instead, it synchronously inputs the deviation ε = W_final − W_t between the actual material weight and the target weight into the adaptive learning model. The goal of the model is to update the filling coefficient k_f and the shape influence coefficient k_s, but the direction of the update needs to be interpretable. In this embodiment, the task-level ε is decomposed into two chains of evidence: if, in multiple tasks with constant θ and ω, the growth rate r of W_c is consistently lower than the theoretical slope derived from ρ and θ, it indicates that the cavity is not effectively filled, which is more likely due to an insufficient opening angle caused by a small k_f or a low k_s value; conversely, if r is too high and the probability of the bridge arch is low, θ is mostly a signal of excessive opening, and k_s should be adjusted down. The adaptive model uses a simple linear corrector with constraint terms. The input is the batch shape statistics (roundness, aspect ratio distribution) and ε, and the output is Δk_f and Δk_s. It is forced to accumulate batch by batch within a small step size to avoid the parameters being pulled out of control by a single anomaly. The intrinsic connection between shape statistics and compensation direction is physical: elongated grains and rough surfaces reduce filling and flow, and parameters need to be adjusted in the direction of increasing opening and extending opening time.

[0058] The technical effects of this embodiment can be summarized on two levels. For process control, the shut-off time is determined by an interpretable weight threshold and growth rate. The three-component compensation system unifies the necessary physical tail differences, reducing the reliance on empirical estimations. For parameter maintenance, the deviation's trajectory is clearly traced back to the filling and shape parameter channels, allowing the same material on the same equipment to gradually converge to more stable control parameters over a long period. Ultimately, the final-state mass distribution within the container more closely matches the task settings, and the entire evidence chain—ΔW, W_comp, t_trig, wake curve, Δk_f / Δk_s—is recorded for verification and reference in subsequent batches.

[0059] As described above, the multi-variety dynamic sampling precision control method provided in this application can accurately identify material varieties through innovative design of a multi-feature fusion model, shape analysis, and spectral detection. A feeding control system is constructed, combined with multiple compensations, to establish a reliable weight monitoring mechanism. Adaptive learning is introduced, and the accuracy of sampling is ensured through parameter updates and predictive shutdown. This method effectively solves the shortcomings of traditional technologies in material identification, feeding control, and shutdown accuracy, providing technical support for dynamic sampling.

[0060] In one embodiment of the method for precise control of dynamic sampling quantity for multiple varieties in this application, the method may further include the following: Step S201: Acquire high-resolution images of grain samples under standard light source illumination, perform Gaussian filtering on the images to eliminate noise, use the Canny operator to perform edge detection to extract particle contours, convert the contours into binary images, perform morphological operations on the binary images to remove edge burrs, perform particle segmentation based on connected component analysis, and calculate the pixel area, boundary perimeter, and roundness index of each particle region after segmentation as shape features. Step S202: Turn on the light source of the near-infrared spectral acquisition device, scan the grain sample to obtain reflectance spectral data, subtract the dark current signal from the reflectance spectral data, perform whiteboard calibration to eliminate instrument drift, use multivariate scattering correction method to remove baseline offset, resample the corrected spectral data according to the preset wavelength interval, and perform standard normal transformation on the resampled data to generate standardized spectral features.

[0061] Optionally, this embodiment focuses on S201 and S202, with the scene set at the sample sorting workbench in the quality inspection room of the acquisition point. In this embodiment, a representative sample cup is gently poured onto an anti-reflective black substrate, spread into a single thin layer, with particles overlapping as little as possible. The sample cup is then pushed into a D65 standard light source box, and an industrial camera is used to capture the image from a vertical direction. To prevent subsequent edge extraction from being hampered by noise and glare, white balance is first performed using a gray card, and exposure and gain are locked for this batch. The original image enters the preprocessing stage. In this embodiment, a small-scale Gaussian filter is used to smooth salt-and-pepper noise while preserving boundary transitions. The kernel width is adjusted using a validation set to avoid smoothing out details such as the abdominal groove. Edge detection is performed using Canny, with two thresholds adaptively set with Otsu's coarse segmentation as a reference. The intersection of the output gradient edge and the brightness threshold partition further suppresses false edges caused by background highlights. The edges are then filled with a binary mask. Considering that some particle boundaries have jagged edges and burrs, we perform an opening operation on the binary image to remove small protrusions, and then perform a closing operation to fill in the outline gaps. This will make the topology of subsequent connected component analysis more stable.

[0062] This embodiment performs step-by-step correction on particle segmentation. Specifically, connected component analysis first calculates the area and bounding rectangle of each region. Slender and large-area connected candidates enter the "concave point segmentation" branch. In this embodiment, concave point detection is used to find deep concave cutting points, combined with the watershed of distance transformation, to avoid dividing two closely adjacent particles into three parts or not dividing them at all. For each segmented region, this embodiment uses the pixel area A and the boundary perimeter P as basic quantities, and simultaneously calculates the roundness C = 4πA / P. 2 (Where A represents the projected area of ​​the grain, P represents the length of the corresponding closed boundary, and C is a dimensionless roundness index, with the value closer to 1 being closer to a circle). The selection of this set of features has its own field logic: the projection of the grain on the shooting plane is approximately preserving the volume, and broken grains will have a smaller A and a relatively larger P, resulting in a decrease in C; the differences between slender grains and varieties will also be reflected in the increase of P relative to A, providing a basis for the subsequent estimation of the shape influence coefficient and filling coefficient.

[0063] In addition to shape feature statistics, this embodiment also includes quality control.

[0064] In this embodiment, the quantiles and tail proportions of the single-particle area distribution are statistically analyzed for each frame. If a sudden increase in the proportion of large-area contiguous particles is detected, it indicates that the material is too densely laid or locally adhered, and the system prompts for re-laying. If the roundness distribution suddenly has two peaks and too many small areas appear simultaneously, it is mostly due to reflection from the inner wall of the light source box or dust adhesion. In this embodiment, a polarizer is added to the image domain or the gradient term weight of the threshold is increased. These processing steps are not for "aesthetics," but to ensure that the shape features can stably reflect the true particle shape and not treat the device status as sample characteristics.

[0065] Optionally, S202 activates the halogen tungsten lamp source of the near-infrared spectroscopy device, and the fiber optic probe performs surface scanning of the sample layer at a fixed working distance to acquire the reflectance spectrum I_s(λ). Within the acquisition frame, the dark current I_dark is first subtracted to eliminate the detector background. Then, calibration is performed using a whiteboard reference I_ref(λ), constructing the reflectance R(λ) = (I_s−I_dark) / (I_ref−I_dark). Baseline fluctuations mainly arise from particle size and scattering differences. In this embodiment, multivariate scattering correction (MSC) or standard normal transformation (SNV) is used for correction. The physical basis of this step is that particle scattering applies approximately linear scaling and translation at each wavelength, creating a false signal that "appears" tilted to the chemical absorption shape. After correction, the relative shape of the absorption band is preserved. Subsequently, R(λ) is resampled at preset wavelength intervals, saturated edge segments are removed, and a fixed-length spectral vector is obtained. A standard normal transformation is then performed to ensure scale consistency between batches, creating comparability for subsequent model input.

[0066] This embodiment addresses both quality control and shape domain considerations for spectral data. In this embodiment, the signal-to-noise ratio (SNR) and drift index are calculated for each scan. If the SNR falls below a threshold, the scan speed is reduced or the averaging frequency is increased. If the baseline after whiteboard calibration still shows slow drift, it indicates that the thermal steady state of the light source has not been achieved, and the system delays entering formal acquisition. This meticulous approach is necessary because the differences in characteristic absorption bands between different varieties are not always on the order of "visually perceptible," and any systematic error will cause the model to hesitate at the boundaries. Through this process, the correlation between the spectrum input to the model and the actual chemical composition more closely resembles physical laws: the absorption intensity of protein-related bands is linked to the concentration of nitrogen-containing groups, while the amylopectin-related bands are associated with starch structure, corresponding one-to-one with the variety statistics stored in the database in this embodiment.

[0067] This embodiment aligns the two feature paths to the sample dimension. The image-side output is a concatenation of single-grain statistics and frame-level statistics for each batch, while the spectral-side output is the standardized spectral vector or its principal component coefficients. The variety identification model will not be detailed here, but its input / output will be explained in this embodiment: the input includes shape statistics (mean, variance, and quantiles of A, P, and C) and spectral feature vectors, and the output is the probability distribution for each candidate variety. The causal relationship between the two types of features and the variety label is closer to natural facts than a "forced fit": for example, indica rice has lower roundness in shape statistics compared to japonica rice, and the local shapes of certain CH and OH combination bands in the near-infrared spectrum are slightly different. After model fusion, a higher probability of indica rice is given, and the database can then extract the corresponding packing density and filling parameters.

[0068] The technical advantage of this embodiment lies in the stable and interpretable extraction of both "visible shape" and "invisible chemistry," and then aligning them within the same batch. S201 separates the particles from adhesion and noise, providing quantities directly related to flow and filling, such as area, perimeter, and roundness. S202 cleans the spectrum from instrument and scattering disturbances, providing a standardized spectrum corresponding to the characteristics of the variety. Both sub-processes are closed-loop, paving the way for downstream parameter retrieval and feed control, avoiding patching up at the next step. For quality inspectors requiring review, this embodiment provides a batch profile, single-particle statistics, spectral shape, and correction logs, reconstructing the original physical state and determining whether the model output conforms to common sense. This traceability is particularly convenient in dispute resolution.

[0069] In one embodiment of the method for precise control of dynamic sampling quantity for multiple varieties in this application, the method may further include the following: Step S301: Concatenate the shape features and standardized spectral features of the grain sample to construct a feature vector, perform principal component analysis to reduce the dimensionality of the feature vector, input the dimensionality-reduced features into a pre-trained convolutional neural network, extract local feature maps, input the feature maps into a fully connected layer to calculate the probability score of each variety, and determine the material variety identifier based on the highest probability score. Step S302: Based on the material type identifier, retrieve the corresponding record in the parameter database, read the basic bulk density value corresponding to the type, multiply the basic bulk density by the temperature and humidity correction coefficient to obtain the real-time bulk density parameter, read the valve filling coefficient value and shape influence coefficient value corresponding to the type, and write the real-time bulk density parameter, filling coefficient value and shape influence coefficient value into the parameter cache area of ​​the control model.

[0070] Optionally, in this embodiment, the acquisition step of S101 and the volume decomposition control of S102 are connected, and S301 and S302 are executed under the condition of continuous material supply from the sampling station. In this embodiment, the single-particle shape statistics (sample mean, standard deviation and quantile of area, perimeter, and roundness) within the previous time window are first aligned with the standardized near-infrared spectrum (reflectance intensity sequence after whiteboard calibration, MSC or SNV, and resampling) according to the sample timestamp to form a spliced ​​feature vector x. Since the shape and spectral dimensions are significantly different, direct parallel connection will cause "axial bias" in the metric space. Therefore, x is first standardized and then PCA is performed to reduce the dimension, retaining the components that can explain the main variance, so as to compress redundancy and improve the focusing ability of subsequent convolution kernels on the effective subspace. PCA does not replace model learning. It extracts the main direction of "shape-spectrum" covariance. For example, the co-occurrence of long-grained increase and absorption enhancement in specific bands is identified as a few principal components, which is beneficial for downstream convolution to capture local patterns.

[0071] In this embodiment, the dimensionality-reduced sequence is input into a pre-trained convolutional neural network. A 1D convolutional kernel on one side of the network slides along the spectral axis to detect narrowband absorption combinations; a smaller convolutional kernel on the other side performs local correlation extraction on the principal component sequence of shape, aiming to identify grouped changes of "decreasing roundness—increasing perimeter." The two feature paths are fused in the middle layer through channel attention. The attention weights are not arbitrary hyperparameters but are driven by acquisition quality indicators: when the spectral signal-to-noise ratio decreases or the image edge quality score is low, the corresponding channel weights are automatically withdrawn, and the model relies more on credible evidence from the other channel. The convolutional output is global pooled to obtain a local feature map z, which is input to a fully connected layer and Softmax to obtain the probability vector p for each variety. The material variety identifier b corresponding to the largest component in p is taken. In this embodiment, the discrimination boundary is constrained to be consistent with natural laws during the training phase. For example, the difference in long grain morphology between indica rice and japonica rice and the difference in absorption in a specific near-infrared band are explicitly included in the label relationship, so that the model does not make decisions based on noise.

[0072] In this embodiment, after obtaining b, the parameter database is retrieved in step S302. The parameter database uses variety-region-season as the primary key, recording the calibration values ​​of the basic bulk density ρ_base, valve filling coefficient k_f, and shape influence coefficient k_s, along with environmental correction rules. The temperature T and relative humidity H from the environmental sensor are read, and ρ_base is multiplied by the temperature and humidity correction coefficient κ(T,H) to obtain the real-time bulk density ρ=ρ_base·κ(T,H). The physical meaning of κ is the equivalent mapping of the influence of water content change and temperature on the bulk porosity in density, and the value comes from the standard sample verification during the acceptance period. Then, k_f and k_s are retrieved from the records, which respectively characterize the effective filling rate of the cavity and the correction of the particle shape on the relationship between the opening angle and the discharge volume. The three parameters are written into the parameter cache area of ​​the control model, with a version number and timestamp, for use in S102 volume decomposition and S103 compensation and shutdown prediction.

[0073] This embodiment considers the stability issues under mixed-flow and abrupt change conditions, and adds an uncertainty judgment layer after the probability output of S301. If the interval between the highest probability and the second highest probability is less than a threshold, or the short window entropy value increases, the system does not immediately switch b completely, but uses sliding interpolation: ρ, k_f, and k_s transition linearly according to the parameters of the previous variety and the candidate variety, and the transition time is linked to the intensity of the incoming material fluctuation. The logic behind this is that the composition of the incoming material does change during the mixed-flow stage, and a hard switch would push the volume decomposition and shutdown towards oscillation. In this embodiment, the image and spectral distribution during this period are also recorded to see if there are double peaks. If the double peaks persist, the operator is prompted to confirm whether the upstream line has merged.

[0074] For example, for the northern hard wheat batch, the spectrum shows stronger absorption in the protein-related band, and the principal components of the shape show higher roundness and shorter perimeter. The network output b = "hard wheat", ρ_base is high, k_f is moderate, and k_s is close to 1. After entering S102, the volume decomposition gives a smaller V_t / cycle, and the shutdown prediction pressure is not significant. Another segment is the mixed flow of early indica rice and broken rice. The shape PC shows a bimodal pattern, and the entropy value of p increases. In this embodiment, the trigger parameter transition is performed. After ρ is corrected according to κ(T,H), it slides between the two sets of ρ_base. k_s is slightly higher, and S102 and S103 can still maintain a stable weight trajectory without jitter. Finally, this S301-S302 link transforms the "visible variety and shape" into "usable density and device parameters" and uses data to support the uncertainty, providing a solid starting point for the number of cycles, angle, time, and compensation downstream.

[0075] In one embodiment of the method for precise control of dynamic sampling quantity for multiple varieties in this application, the method may further include the following: Step S401: Calculate the target feeding volume by the ratio of the target feeding weight to the real-time bulk density parameter. Divide the target feeding volume by the product of the number of valve cells and the volume of a single cell to obtain the theoretical feeding volume of a single valve turn. Calculate the actual feeding volume of a single valve turn based on the product of the theoretical feeding volume of a single valve turn, the filling coefficient parameter, and the shape influence coefficient parameter. Divide the target feeding volume by the actual feeding volume of a single valve turn and round down to obtain the integer number of turns. Step S402: Subtract the product of the integer number of turns and the actual discharge volume per turn of the rotary valve from the target discharge volume to obtain the remaining volume. Divide the remaining volume by the volume of a single cell to calculate the number of cells to be opened. Multiply the number of cells to be opened by the angle corresponding to a single cell to obtain the theoretical opening angle. Multiply the theoretical opening angle by the shape influence coefficient parameter to calculate the final opening angle.

[0076] Optionally, this embodiment focuses on steps S401 and S402, with the scenario being the calculation of the rotary valve's movement within an online sampler before grain enters the warehouse. In step S101, the grain variety has been identified, and the real-time bulk density ρ, filling coefficient k_f, and shape influence coefficient k_s have been retrieved from the parameter library. The target discharge weight W_t is given by the task order. The control calculation unit first projects W_t into the volume domain to obtain the target discharge volume V_t = W_t / ρ. The reason for this volume-based calculation is straightforward: the rotary valve discharges based on the compartment volume. If the number of rotations and angles were directly calculated in the mass domain, density differences would be mixed into the mechanical movements, making it inexplicable at the equipment level.

[0077] This embodiment then imports the geometric parameters from the equipment file: the number of rotary valve cells N_cell and the nominal volume of a single cell V_cell, with the theoretical discharge volume per turn V_theo = N_cell·V_cell. Actual discharge is lowered by two factors: insufficient filling rate, attributed to k_f; and the influence of particle shape on flow and bridging, attributed to k_s. Under the combined effect of these two factors, the actual discharge volume per turn of the rotary valve V_rev = V_theo·k_f·k_s. In this embodiment, V_t is divided by V_rev and rounded down to obtain the integer number of turns n_rev. This "rounding down" is for conservatism—the excess volume is compensated for by subsequent angles, avoiding excessive turns that lead to overshoot and difficulty in braking. For example, indica rice is slender, and glutinous corn is stickier, resulting in smaller k_s or k_f, which in turn reduces V_rev, naturally lowering n_rev, and consequently increasing the size of subsequent angles.

[0078] This embodiment continues to process the remaining volume in S402. First, the difference is obtained using V_rem = V_t − n_rev · V_rev, and then this part is converted into the number of cells to be opened: n_open = V_rem / V_cell. Theoretically, n_open doesn't have to be an integer, but mechanically it can only be incremented by cells. In this embodiment, n_open is split into an integer part and a fractional part. The integer part is directly mapped to an angle step, with the angle corresponding to a single cell being 360° / N_cell. The insufficient amount caused by the fractional part will be compensated at the end by extending the opening time or fine-tuning the rotation speed to reduce angle quantization error. Thus, the theoretical opening angle θ_th = n_open · (360° / N_cell) is obtained. Considering that the shape influence not only affects filling but also changes the slope of the "opening angle - instantaneous flow rate" curve, in this embodiment, θ_th is multiplied by k_s to obtain the final opening angle θ = θ_th · k_s. k_s>1 is commonly seen in slender grains, which is equivalent to giving a larger angle for the same target volume to counteract the tendency to form arches in the cavities; if the model estimates k_s to be slightly less than 1, it usually corresponds to short and round varieties, where the cavities can be filled more fully and the angle does not need to be too large.

[0079] In this embodiment, if V_rem is very small, smaller than a certain threshold of the single-cell volume, a small cell is not forcibly opened. Instead, this small volume is handled by the compensation loop of S103 using fine-tuning of the opening time to avoid the mechanical action time cost exceeding the benefit. If V_rem is abnormally large, and close to more than half of V_rev, in this embodiment, it is preferable to add one revolution to n_rev and then decrease θ. This way, adding a small angle with each revolution is more controllable in terms of torque fluctuation and braking consistency than a large angle with a single revolution. The rationale behind these two trade-offs is that the nonlinearity of motor inertia and braking response is more pronounced at large angles, and distributing it across revolutions is more conducive to the convergence of the subsequent compensation closed loop.

[0080] This embodiment illustrates how the parameters are implemented for two specific batches. First, the early indica rice batch: ρ is slightly lower, the roundness distribution is smaller, the model gives k_s>1, the calculated n_rev is lower, and θ is higher. After one round of splitting, the equipment completes the target volume using a combination of "multiple small angles." Second, the autumn japonica rice batch: ρ is higher and the shape is rounder, k_f and k_s are close to 1, n_rev is larger, and θ is smaller, with the movement closer to ideal geometry. The difference between the two is reflected in subsequent real-time monitoring: the W_c curve of the former relies more on the completion of angle segments, while the latter is mainly contributed by the number of rotations.

[0081] This embodiment also adds a practical layer of error protection. After initial equipment installation or seasonal changes, the nominal values ​​of V_cell and N_cell may deviate slightly from the actual values. In this embodiment, during small-batch trial runs, the W_c increment for each "circle movement" is recorded, and the actual V_rev is deduced to serve as a temporary overlay value for this shift. This value is then written back to the parameter library once sufficient evidence has been accumulated through adaptive learning. The deduction is based on mass conservation and the reliability of ρ, which comes from the S101 spectral-variety link and is more stable than the mechanical nominal value in the short term.

[0082] The technical advantages of this embodiment are reflected in two layers: First, it accurately converts the quality target into a mechanical quantity of "number of revolutions + angle," handling the full step and fine-tuning separately to facilitate consistency at the execution end. Second, it aligns the parameters with the physical phenomena, preventing k_f and k_s from being mystified. They respectively handle the reduction of volume and amplification of angle along the filling and shape paths, which can be explained and corrected by subsequent closed-loop adjustments. This link from W_t to n_rev and θ is clearly traceable, providing a stable starting point for the compensation and braking of S103.

[0083] In one embodiment of the method for precise control of dynamic sampling quantity for multiple varieties in this application, the method may further include the following: Step S501: Divide the final opening angle by the rotation speed of the rotary valve to calculate the basic rotation time. Take the sum of the basic rotation time and the material falling time as the theoretical opening time. Add the sum of the sensor response delay time and the control system delay time to the theoretical opening time to obtain the actual opening time. Set the operating parameters of the servo motor and the start-up sequence of the electromagnetic braking mechanism according to the actual opening time. Step S502: Read the material level height data collected by the multi-layer material level sensor array, calculate the distance between each sensor position and the current material level height, assign sensor weight coefficients according to the distance, sum the product of the material level height data and the weight coefficients to obtain the weighted average material level height, and calculate the current material discharge weight by multiplying the weighted average material level height by the bottom area of ​​the receiving device, the real-time bulk density parameter, and the material surface distribution correction coefficient.

[0084] Optionally, this embodiment focuses on steps S501 and S502, with the working scenario being the sample sorting station performing a target unloading task under continuous feeding conditions. In step S102, the final opening angle θ and the rotary valve rotation speed ω have been calculated, and the t_fall (material fall time) and system delay parameters have been calibrated. The control unit first uses θ and ω to derive the basic rotation time t_rot = θ / ω, and adds t_rot and t_fall to obtain the theoretical opening time t_th. Since the weight closed-loop depends on the actual response of the sensor link, in this embodiment, the sensor response delay t_sens and the control system delay t_ctrl are superimposed to calculate the actual opening time t_open = t_th + t_sens + t_ctrl. t_fall is mapped from the empirical curve of the drop distance h and particle air resistance, and will increase moderately when the batch is wet or dusty; t_sens comes from the dynamic response test of the material level array, and t_ctrl is composed of the PLC cycle and driver delay, both maintained in the equipment file. After obtaining t_open, the PLC sets the servo motor's speed profile to a three-stage sequence of "constant speed - deceleration - braking": the first stage rapidly approaches 80% of θ with the set ω; the second stage decelerates near the shut-off window; and the third stage triggers the braking coil at t_open−t_b (t_b is the electromagnetic braking setup time) to ensure the valve plate closes near the theoretical stopping point. If θ is large and ω is high, the deceleration stage will be lengthened to avoid overshoot due to inertia. This timing arrangement is consistent with the rotational inertia constraint of the mechanical system.

[0085] This embodiment does not assume that t_fall is stable. To avoid systematic errors caused by seasonal changes, this embodiment performs short-range calibration with standard sand after startup: keeping θ and ω constant, the delay inflection point of the W_c curve is measured, and the observed inflection point time is compared with the estimated value to correct t_fall. The correction is not hardcoded globally, but is attached to a temporary entry under the current shift and current humidity range to prevent a single accidental wind speed change from "tying up" the entire process. In addition, if the airflow at the top of the container is detected to be enhanced by the dust collector during the mission, this embodiment fine-tunes t_fall based on the reading of the top wind speed sensor and simultaneously adjusts t_open. The logic is that the higher the wind speed, the greater the equivalent resistance, and the slightly longer the fall time.

[0086] In this embodiment, the material level height data h_i from the multi-layer material level sensor array is read in step S502. The array is distributed along the container wall from bottom to top, and the installation angle avoids the direction of material impact. The control unit calculates the distance d_i from each sensor to the material surface based on the estimated current material level height in the container and the geometric position of each sensor. Sensors with a small d_i are more susceptible to splashing and adhesion, and are assigned a lower weight. Sensors located in the middle section have a slightly higher weight. The weighting coefficient w_i is generated by a rule that varies with the half-apex angle φ of the stack cone and the wall friction state μ. φ and μ are obtained by fitting historical stack shapes. The weighted average material level height is obtained as h̄=Σ(w_i·h_i) / Σw_i. To reduce the deviation caused by single-point failure, in this embodiment, h_i is subjected to box filtering and median removal before summation to remove extreme values ​​before weighting.

[0087] In this embodiment, when converting h̄ to the current feed weight W_c, container geometry and stack shape corrections are considered. The bottom area A of the receiving device is given by the CAD model, and the material surface distribution correction coefficient γ is used to map the actual volume of the "frustum / cone mixed stack shape" to the "cylindrical approximation" volume. γ is not set arbitrarily, but is a function based on the half-apex angle φ of the stack cone. The smaller φ is (the more pointed the stack), the smaller the actual volume under the same h̄, and the smaller γ is. The real-time bulk density ρ is output by S302, so W_c = ρ·A·h̄·γ. In this expression, ρ represents the bulk density of the current batch of material, A is the effective bottom area of ​​the container, h̄ is the weighted average material level height, γ is the stack shape correction coefficient, and W_c is the currently estimated feed weight. In this embodiment, W_c is smoothed over time using a small-window Kalman filter, the process model uses the mass growth rate derived from the previous round of θ, ω, and ρ, and the observation noise is estimated using the sensor's historical variance, thereby reducing instantaneous jitter without sacrificing response.

[0088] This embodiment monitors W_c and angle progress together, establishing a three-dimensional trajectory of "angle-time-mass" to identify abnormal flow patterns. If the upper sensor h_i is detected to decrease rapidly while the lower sensor remains almost still, it is initially judged as the expansion of an arch or void. The system briefly increases ω and vibrates the valve body briefly to break the arch before returning to the original setting. If the overall h_i decreases slowly but W_c increases very slowly, it is judged as "material surface slippage" based on wind speed and powder content. At this time, γ is appropriately lowered to acknowledge the reality of a sharper pile shape, making the mass estimation closer to reality.

[0089] This embodiment incorporates adaptive weighting. Each sensor in the material level array has a long-term health score. If a sensor has a high residual adhesion ratio in the last three tasks, its weight will be reduced by one level from the initial value until it recovers after maintenance and cleaning. For high-humidity corn batches, this embodiment observed that the upper ultrasonic sensors are more susceptible to interference from atomized water vapor, and the weighting rules will be biased towards the lower layers.

[0090] This embodiment combines the timing results of S501 with the weight estimation of S502. During the t_open countdown, the control unit updates W_c every Δt and compares it with the target W_t, calculating the real-time difference and growth rate as input to the compensation model in the next step, S103. If the growth slope of W_c deviates significantly from the theoretical slope derived from θ and ω in the latter half of the countdown, the system prioritizes temporary corrections to γ ​​and t_fall instead of immediately recalculating θ, avoiding large actions at the end stage that could cause oscillations.

[0091] This embodiment provides two specific scenarios to facilitate understanding of the logic chain. First, in batches of long-grain indica rice, θ is larger, ω is not set high enough, t_rot is longer, and the pile shape is more pointed. In this embodiment... W_c estimation avoids systemic overestimation; secondly, for wet batches of waxy corn, t_fall is initially too high, and h_i fluctuates due to moisture. In this embodiment, after startup calibration, t_fall is increased by a small step, and the noise in the filtering process is also increased slightly. As a result, the W_c curve is smoother, and the lead time for shutdown prediction is more controllable. Both approaches aim to connect the "material properties-geometry-time" chain, preventing any unexpected event at one end from causing the entire system to deviate.

[0092] The technical advantages of this embodiment are reflected in three aspects. First, the calculation of t_open clearly distinguishes several inevitable time terms, eliminating reliance on experience for the timing of servo and electromagnetic braking, and properly placing mechanical inertia within the deceleration phase. Second, the weighting of multiple material levels and the online correction of γ make W_c closer to the actual stockpile, avoiding misjudgment of quality due to imperfect stockpile shape. Third, the ternary trajectory of W_c—θ—ω provides clean input for subsequent compensation and turn-off time prediction in S103, and also provides an interpretable record of deviation sources for long-term adaptive testing. When reviewing for quality inspection or maintenance, the curves of θ, ω, t_open, each h_i with weights, and γ with W_c can be pulled out and compared to quickly pinpoint whether the deviation originates from time delay, stockpile shape, or sensor status.

[0093] In one embodiment of the method for precise control of dynamic sampling quantity for multiple varieties in this application, the method may further include the following: Step S601: Subtract the target material weight from the current material weight to obtain the weight difference. Set a compensation coefficient based on the weight difference. Multiply the product of the valve opening area and the drop height by the real-time bulk density parameter to obtain the air material compensation weight. Multiply the product of the material flow rate and the system response time by the real-time bulk density parameter to obtain the system delay compensation weight. Multiply the product of the brake overshoot volume and the real-time bulk density parameter to obtain the brake overshoot compensation weight. Add the air material compensation weight, the system delay compensation weight, and the brake overshoot compensation weight to obtain the total compensation weight. Step S602: Collect the material feeding weight data of two adjacent sampling periods, calculate the difference between the material feeding weight data and divide it by the sampling time interval to obtain the weight growth rate, subtract the total compensation weight from the target material feeding weight to obtain the shutdown threshold, multiply the shutdown threshold by the system safety factor and subtract the minimum material feeding weight threshold to obtain the actual shutdown threshold, and calculate the time required for the current material feeding weight to reach the actual shutdown threshold based on the weight growth rate.

[0094] Optionally, after the initial volume-angle-time calculation given in S102, this embodiment proceeds to S601 and S602 to quantify the early shut-off time using closed-loop quantification. In this embodiment, the current discharge weight W_c is obtained from the mass estimation loop of the receiving device, and the difference between it and the target discharge weight W_t set in the task is used to obtain ΔW=W_t−W_c. This difference is only superficial; the closure of the valve does not mean that the material immediately stops flowing in. Particles in transit within the drop distance, control link lag, and the inertia of the valve plate and drive will all adjust the final reading upwards.

[0095] In this embodiment, three types of compensation are constructed according to physical decomposition in S601. The airborne material compensation weight W_air is used to estimate the mass of particles in transit between the valve orifice and the container. Based on the current valve opening area S and the equivalent fall height h, the in-transit volume under steady-state flow is approximately the product of S and the characteristic fall time, which is multiplied by the real-time bulk density ρ to obtain W_air. The system delay compensation weight W_lat is characterized by the material flow rate v_flow and the system response time t_sys. During the time from the issuance of the braking command to the actual reduction of the valve gap, there is still a volume of material of v_flow·t_sys that must pass through, which is also multiplied by ρ. The braking overshoot compensation weight W_ov comes from the brake overshoot volume V_ov, which is the equivalent "residual opening" volume formed by the valve plate due to inertia and rebound during the braking phase. It is multiplied by ρ to obtain the mass expression. The three compensations are obtained independently and finally linearly added to obtain the total compensation weight W_comp = W_air + W_lat + W_ov. To avoid arbitrary assumptions based on experience, S and v_flow are derived from the instantaneous mapping of encoder angle and historical calibration, h is derived from the survey calibration, t_sys is derived from PLC cycle, driver response and relay time, and V_ov is derived from the short-range test run fitting curve before the shift.

[0096] This embodiment considers the compensation coefficient to adjust automatically according to operating conditions. In this embodiment, the compensation coefficient α is set by the sign and amplitude of ΔW, and the nominal values ​​of W_air and W_ov are slightly scaled to adapt to the extended wake caused by dust adhering to the wall or wet batch "stringing". However, W_lat is left unchanged because the control link lag is related to the equipment status rather than the material properties. The adjustment only takes effect within a short window and returns to the nominal value after the window ends to avoid treating one-time adhesion as a persistent characteristic. This distinction follows natural laws: air-material interaction and wall adhesion mainly affect in-transit and overshoot, while the control delay is more "electronic" and does not jump with the grain.

[0097] In this embodiment, before estimating the shutdown timing in S602, the mass growth rate r is first evaluated. The difference between W_c of two adjacent sampling periods is divided by the sampling interval Δt to obtain the instantaneous r = dW / dt. To suppress jitter, an exponential weighting is introduced to maintain short-term sensitivity. Subsequently, the target feed weight is subtracted from the total compensation to obtain the mass-level shutdown threshold W_th = W_t − W_comp. For equipment and safety redundancy considerations, this embodiment introduces a system safety factor K_s and a minimum feed weight threshold W_min. The former is to prevent premature braking caused by short transient distortion, and the latter is mapped to the lower limit of "no further pursuit of insignificant sporadic feed". The final actual shutdown threshold is W_th,act = K_s·W_th − W_min. From this, the remaining time t_hit = (W_th,act − W_c) / r to reach the threshold can be calculated, and the trigger time is arranged accordingly to offset the combined lag of electromagnetic braking and descent.

[0098] This embodiment incorporates two robustness measures in the computational chain. First, when r drops sharply due to a short-term material shortage upstream, t_hit might be exaggerated. In this embodiment, considering the upstream material level or flow meter status, if a "material shortage" is determined, the system switches to a minimum early trigger strategy, executing a combination of guaranteed t_b, t_sys, and t_fall to prevent task delays caused by prolonged waiting. Second, if ΔW is negative and its amplitude increases, it indicates that W_c has exceeded W_t. The system increases the empirical term of W_ov and records an "overshoot event." The next batch will start with a more conservative V_ov to avoid repeating the same mistake. The basis for these two strategies is not mysterious; both are derived from the causal relationship between material and equipment: when there is a material shortage, the rate is unusable; when overshoots occur frequently, inertia or adhesion is excessive.

[0099] In this embodiment, the angle-speed plan of S102 is aligned with the current t_hit. If t_hit is less than the remaining time required for the action according to θ and ω, it means that the existing angle is insufficient to fill the target before the threshold. The system will briefly increase ω or slightly increase the θ margin without changing n_rev. Conversely, if t_hit is much greater than the action time, it means that the current flow rate is too high, and ω will be reduced earlier to pull back the growth slope. The adjustment range is limited by mechanical and safety boundaries and will not exceed the maximum angle and speed set in S102.

[0100] In this embodiment, at the end of the task, the final error ε = W_final − W_t is recorded and fed into the adaptive learning model. Besides ε, the model input also includes the shape statistics (roundness, aspect ratio distribution) and the proportion of the three compensation components for that batch. The output is small suggested values ​​for Δk_f and Δk_s, but these are only fixed into the parameter library after multiple batches with the same directional deviation are confirmed. To explain this in layman's terms: if the deviation is consistently too light, and always occurs when W_air and W_ov are small, it's most likely due to insufficient filling caused by shape, and k_s should be gradually increased; if the deviation is more related to the proportion of W_lat, check the control link rather than the material parameters. This cause-and-effect graph brings the conclusions learned by the machine back to physical relationships that humans can understand.

[0101] The technical effect of this embodiment converges on a simple goal: to make shutdown independent of luck. Three types of compensation include unavoidable tail quality in a reliable manner, the coordination of r and W_th,act provides a specific timing schedule, and the safety factor and minimum threshold prevent unforeseen problems in boundary conditions. Combined with the feedforward in S102 and the learning after S103, the entire link is physically closed and data-traceable, suitable for long-term operation at the acquisition site without relying on the "feel" of skilled workers.

[0102] In one embodiment of the method for precise control of dynamic sampling quantity for multiple varieties in this application, the method may further include the following: Step S701: The predicted weight is obtained by adding the product of the weight growth rate and the time required to reach the shut-off threshold to the current feeding weight. The early shut-off time is calculated based on the predicted weight. The electromagnetic braking time, the control system delay time, and the material falling time are subtracted from the early shut-off time to obtain the actual shut-off trigger time. At the actual shut-off trigger time, the servo motor is stopped and the electromagnetic brake is activated to shut off the feeding port. The feeding distributor is driven to transport the feeding material to the bulk density measuring container, the moisture measuring container, and the imperfect particle measuring container respectively. Step S702: Subtract the target material weight from the actual material weight to obtain the weight deviation, calculate the ratio of the weight deviation to the target material weight to obtain the relative error, calculate the filling coefficient correction amount and the shape influence coefficient correction amount based on the relative error, subtract the filling coefficient correction amount from the original filling coefficient parameter to obtain the updated filling coefficient parameter, and subtract the shape influence coefficient correction amount from the original shape influence coefficient parameter to obtain the updated shape influence coefficient parameter.

[0103] Optionally, this embodiment builds upon the weight growth rate and actual shutdown threshold obtained in S601 / S602, and focuses on the end-of-line convergence control and parameter self-learning of S701 and S702 in the sampling task of the acquisition line.

[0104] In this embodiment, the current feed weight W_c, the weight growth rate r, and the time t_hit required to reach the shutdown threshold are first considered as a set of time-domain quantities for forward looking, predicting the weight evolution without intervention. The predicted weight is obtained by linear extrapolation W_pred = W_c + r·t_hit. The linear assumption is not arbitrary; it is based on the stationarity of r within the short window and the approximate constant current characteristics after S102 locks θ and ω. If r has a trend of drift within the short window, in this embodiment, exponential weighting is applied when calculating t_hit, so the forward looking can still be close to reality. Based on this, the control unit converts the relationship between W_pred and the shutdown threshold into the advance shutdown time t_adv, that is, "how far in advance should the braking link command be issued from now on", so as to offset the subsequent series of physical and control lags.

[0105] In this embodiment, t_adv is further subtracted from three unavoidable delays to calculate the actual shutdown trigger time t_trig = t_adv − (t_b + t_ctrl + t_fall). Here, t_b is the setup time of the electromagnetic brake from power-on to generating effective braking force; t_ctrl is the total delay of PLC sampling, task scheduling, and driver response; and t_fall is the material fall time caused by the free fall distance from the valve port to the feeder. These three have different physical meanings and should not be confused: t_b depends on the temperature state of the solenoid valve and coil; t_ctrl is related to the control cycle and communication link congestion; and t_fall is affected by the fall distance, particle size, and air resistance, and is slightly related to the batch's humidity and powder content. Upon reaching t_trig, the PLC issues two types of actions: first, it stops the servo motor according to the preset deceleration profile and maintains torque lock; second, it drives the electromagnetic brake to close the valve plate, ensuring rapid narrowing of the valve gap and interrupting the material flow near the threshold. During the same time window when the valve is closed, the feed distributor guides the tail section material to the three measuring containers for bulk density, moisture, and imperfect particles according to the channel mapping of this batch of work order. Interlocking and anti-crosstalk signals are set between the distributor gates to avoid mixed loading caused by two paths opening simultaneously.

[0106] In this embodiment, if a significant decrease in r and a low material level indication from the upstream flow meter are observed within the last two sampling periods before t_trig, the system will reduce the impact of the t_fall term and converge t_trig towards the current moment to avoid premature braking in a material-free state that slows down the cycle time. On the other hand, if r rises within the short window and the material level array shows a rapid decrease in the upper layer while the lower layer has not followed, this embodiment indicates that a surge may have occurred due to a temporary bridge breaking. The equivalent impact of t_b will be weakened, and t_trig will be moderately advanced to cover the tail of the surge. Both fine-tuning methods follow the natural evolution of particle flow and only take effect within the current task to avoid disrupting the baseline for the next iteration.

[0107] In this embodiment, parameter self-learning is performed in step S702 after the task is completed. The system obtains the final actual material weight W_final from the data bus, calculates the weight deviation ε=W_final−W_t, and then calculates the relative error η=ε / W_t. η is a condensed expression of the combined effect of the equipment and the material. In this embodiment, η is not directly proportionally allocated to any parameter, but is solved in a constrained manner based on the evidence chain of this batch. The evidence chain includes: shape statistics (roundness, aspect ratio distribution), θ and n_rev at S102, the proportions of the three components W_air, W_lat, and W_ov decomposed in S601, and the slope trend of r in the final segment. Based on this information, the adaptive learning model outputs the filling coefficient correction Δk_f and the shape influence coefficient correction Δk_s. The update rule adopts small step size and boundary constraints to keep k_f and k_s changing within their respective physical reasonable ranges.

[0108] In explaining the intrinsic relationship between indicators and parameter correction directions in this embodiment, if η is consistently positive and the W_ov ratio is strongly correlated with ω, it indicates that the overshoot caused by inertia and adhesion is too large. A more reasonable approach is to first correct the V_ov mapping of S601, and then observe whether the overshoot problem is alleviated but still too severe before setting Δk_s to a negative value to reduce subsequent θ. If η is negative and the steady-state slope of r is lower than the theoretical slope derived from ρ and θ in multiple batches, while shape statistics show that the average roundness is low and the aspect ratio is high, it indicates that the cavity is not effectively filled. Δk_f and Δk_s should be positive, and the next batch should be given a larger effective opening and more sufficient filling time under the same volume. In this embodiment, the hysteresis of the control link is not mistaken for material parameters. If W_lat is too large in this batch, the t_ctrl path is checked first. If necessary, k_f / k_s is not updated, only the t_ctrl estimate in the control parameter library is updated.

[0109] This embodiment provides a concise update formula, making it easy for implementers to put into practice: Δk_s=β_s·η·ψ_s, Δk_f=β_f·η·ψ_f.

[0110] In the formula, β_s and β_f are the learning step size, η is the relative error, ψ_s is the direction factor composed of shape statistics and the deviation of the final slope, and ψ_f is the direction factor composed of evidence of insufficient filling (e.g., r is lower than the theoretical slope, bridge arch frequency). The values ​​of ψ_s and ψ_f are limited to [−1,1], which physically means mapping the confidence level of "whether the deviation comes from shape influence or filling state" into sign and amplitude, avoiding large adjustments when the evidence is insufficient. The updated parameters are entered into the parameter cache of the control model and archived with batch timestamps. Only when multiple consecutive batches have the same direction under similar varieties and environments are they solidified into the parameter database.

[0111] The technical effects of this embodiment are reflected in two dimensions. In the operational dimension, through precise calculation of t_trig and real-time fine-tuning of nearby anomalies, the shutdown execution shifts from "waiting based on experience" to "advancing based on evidence," preventing the accumulation of systematic deviations caused by early or late shutdown. In the parameter dimension, η is not crudely assigned, but rather linked to a causal chain of W_air / W_lat / W_ov and shape statistics. The update directions of Δk_f and Δk_s can be reviewed and interpreted afterward, suitable for long-term operation and shift handover communication. Ultimately, the sample quality obtained from the three branches of bulk density, moisture, and imperfect particles closely matches the set target, the downstream measurement schedule is not disrupted, and the shutdown log, error, correction amount, and learning step size are all recorded, allowing for a review of the specific physical and parameter states at any given moment in case of disputes.

[0112] To effectively address the shortcomings of traditional technologies in material identification, feeding control, and shut-off accuracy, and to provide technical support for dynamic sampling, this application provides an embodiment of a multi-variety dynamic sampling quantity precision control device for implementing all or part of the aforementioned multi-variety dynamic sampling quantity precision control method. See [link to embodiment]. Figure 2 The multi-variety dynamic sampling quantity precision control device specifically includes the following components: The feeding model determination module 10 is used to collect grain images and near-infrared spectral data, perform edge extraction and shape analysis on the image data, extract the area, perimeter and roundness features of the grain particles, perform baseline correction and standardization processing on the near-infrared spectral data, input the shape features and spectral features into the variety identification model, identify the current material variety, retrieve the corresponding bulk density parameters, filling coefficient parameters and shape influence coefficient parameters from the parameter database according to the material variety, and input the parameters and target feeding weight into the feeding control model. The feeding operation control module 20 is used to calculate the target feeding volume based on the bulk density parameter, divide the target feeding volume by the feeding volume of a single turn of the rotary valve to obtain the integer number of turns, calculate the remaining volume, calculate the final opening angle according to the filling coefficient parameter and the shape influence coefficient parameter, input the opening angle and the rotation speed parameter into the terminal control model to calculate the opening time, and simultaneously collect data from multi-layer material level sensors, calculate the current feeding weight according to the material level data and the bulk density parameter, and monitor the feeding weight in real time. The adaptive shutdown module 30 is used to input the difference between the current feeding weight and the target feeding weight into the compensation model, calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. The shutdown threshold is obtained by subtracting the compensation weight from the target feeding weight. The weight growth rate is calculated, and the time when the shutdown threshold is reached is predicted based on the weight growth rate. The electromagnetic braking mechanism is activated in advance to shut off the feeding port, and the material is introduced into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. At the same time, the deviation between the actual feeding weight and the target weight is transmitted to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

[0113] As described above, the multi-variety dynamic sampling precision control device provided in this application can accurately identify material varieties through innovative design of a multi-feature fusion model, shape analysis, and spectral detection. It constructs a feeding control system, combines multiple compensations, and establishes a reliable weight monitoring mechanism. Adaptive learning is introduced, and the accuracy of sampling is ensured through parameter updates and predictive shutdown. This method effectively solves the shortcomings of traditional technologies in material identification, feeding control, and shutdown accuracy, providing technical support for dynamic sampling.

[0114] From a hardware perspective, in order to effectively address the shortcomings of traditional technologies in material identification, feeding control, and shut-off accuracy, and to provide technical support for dynamic sampling, this application provides an embodiment of an electronic device for implementing all or part of the aforementioned method for precise control of dynamic sampling quantity for multiple product varieties. The electronic device specifically includes the following components: The system comprises a processor, memory, a communication interface, and a bus; wherein the processor, memory, and communication interface communicate with each other via the bus; the communication interface is used to realize information transmission between the multi-variety dynamic sample quantity precision control device and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the multi-variety dynamic sample quantity precision control method and the multi-variety dynamic sample quantity precision control device in the embodiments, the contents of which are incorporated herein, and repeated details will not be described again.

[0115] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0116] In practical applications, parts of the method for precise control of dynamic sample quantity for multiple varieties can be executed on the electronic device side as described above, or all operations can be completed in the client device. The specific choice depends on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0117] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0118] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0119] In one embodiment, the function of the multi-variety dynamic sampling quantity precise control method can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control: Step S101: Collect grain images and near-infrared spectral data, perform edge extraction and shape analysis on the image data, extract the area, perimeter, and roundness features of the grain particles, perform baseline correction and standardization on the near-infrared spectral data, input the shape features and spectral features into the variety identification model, identify the current material variety, retrieve the corresponding bulk density parameters, filling coefficient parameters, and shape influence coefficient parameters from the parameter database according to the material variety, and input the parameters and target feeding weight into the feeding control model; Step S102: Calculate the target feeding volume based on the bulk density parameter, divide the target feeding volume by the feeding volume per single turn of the rotary valve to obtain the integer number of turns, calculate the remaining volume, calculate the final opening angle based on the filling coefficient parameter and the shape influence coefficient parameter, input the opening angle and the rotation speed parameter into the terminal control model to calculate the opening time, and simultaneously collect data from the multi-layer material level sensor, calculate the current feeding weight based on the material level data and the bulk density parameter, and monitor the feeding weight in real time; Step S103: Input the difference between the current feeding weight and the target feeding weight into the compensation model, calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. Subtract the compensation weight from the target feeding weight to obtain the shut-off threshold. Calculate the weight growth rate. Predict the time when the shut-off threshold will be reached based on the weight growth rate. Activate the electromagnetic braking mechanism in advance to shut off the feeding port. Import the material into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, transmit the deviation between the actual feeding weight and the target weight to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

[0120] As described above, the electronic device provided in this application, through innovative design of a multi-feature fusion model, achieves accurate identification of material types via shape analysis and spectral detection. A feeding control system is constructed, combined with multiple compensations, to establish a reliable weight monitoring mechanism. Adaptive learning is introduced, and through parameter updates and predictive shutdown, the accuracy of sample separation is ensured. This method effectively solves the shortcomings of traditional technologies in material identification, feeding control, and shutdown accuracy, providing technical support for dynamic sample separation.

[0121] In another embodiment, the multi-variety dynamic sampling quantity precise control device can be configured separately from the central processing unit 9100. For example, the multi-variety dynamic sampling quantity precise control device can be configured as a chip connected to the central processing unit 9100, and the multi-variety dynamic sampling quantity precise control method function can be realized through the control of the central processing unit.

[0122] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technology.

[0123] like Figure 3As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0124] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0125] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0126] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0127] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device for communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0128] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0129] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored audio via the speaker 9131.

[0130] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the multi-variety dynamic sample quantity precise control method with the execution subject being a server or client in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the multi-variety dynamic sample quantity precise control method with the execution subject being a server or client in the above embodiments. For example, when the processor executes the computer program, it implements the following steps: Step S101: Collect grain images and near-infrared spectral data, perform edge extraction and shape analysis on the image data, extract the area, perimeter, and roundness features of the grain particles, perform baseline correction and standardization on the near-infrared spectral data, input the shape features and spectral features into the variety identification model, identify the current material variety, retrieve the corresponding bulk density parameters, filling coefficient parameters, and shape influence coefficient parameters from the parameter database according to the material variety, and input the parameters and target feeding weight into the feeding control model; Step S102: Calculate the target feeding volume based on the bulk density parameter, divide the target feeding volume by the feeding volume per single turn of the rotary valve to obtain the integer number of turns, calculate the remaining volume, calculate the final opening angle based on the filling coefficient parameter and the shape influence coefficient parameter, input the opening angle and the rotation speed parameter into the terminal control model to calculate the opening time, and simultaneously collect data from the multi-layer material level sensor, calculate the current feeding weight based on the material level data and the bulk density parameter, and monitor the feeding weight in real time; Step S103: Input the difference between the current feeding weight and the target feeding weight into the compensation model, calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. Subtract the compensation weight from the target feeding weight to obtain the shut-off threshold. Calculate the weight growth rate. Predict the time when the shut-off threshold will be reached based on the weight growth rate. Activate the electromagnetic braking mechanism in advance to shut off the feeding port. Import the material into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, transmit the deviation between the actual feeding weight and the target weight to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

[0131] As described above, the computer-readable storage medium provided in this application, through an innovative design of a multi-feature fusion model, achieves accurate identification of material types via shape analysis and spectral detection. A material feeding control system is constructed, combined with multiple compensations, to establish a reliable weight monitoring mechanism. Adaptive learning is introduced, and through parameter updates and predictive shutdown, the accuracy of sample division is ensured. This method effectively solves the shortcomings of traditional technologies in material identification, material feeding control, and shutdown accuracy, providing technical support for dynamic sample division.

[0132] Embodiments of this application also provide a computer program product capable of implementing all steps in the multi-variety dynamic sample quantity precise control method with the execution subject being a server or client in the above embodiments. When this computer program / instruction is executed by a processor, it implements the steps of the multi-variety dynamic sample quantity precise control method. For example, the computer program / instruction implements the following steps: Step S101: Collect grain images and near-infrared spectral data, perform edge extraction and shape analysis on the image data, extract the area, perimeter, and roundness features of the grain particles, perform baseline correction and standardization on the near-infrared spectral data, input the shape features and spectral features into the variety identification model, identify the current material variety, retrieve the corresponding bulk density parameters, filling coefficient parameters, and shape influence coefficient parameters from the parameter database according to the material variety, and input the parameters and target feeding weight into the feeding control model; Step S102: Calculate the target feeding volume based on the bulk density parameter, divide the target feeding volume by the feeding volume per single turn of the rotary valve to obtain the integer number of turns, calculate the remaining volume, calculate the final opening angle based on the filling coefficient parameter and the shape influence coefficient parameter, input the opening angle and the rotation speed parameter into the terminal control model to calculate the opening time, and simultaneously collect data from the multi-layer material level sensor, calculate the current feeding weight based on the material level data and the bulk density parameter, and monitor the feeding weight in real time; Step S103: Input the difference between the current feeding weight and the target feeding weight into the compensation model, calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. Subtract the compensation weight from the target feeding weight to obtain the shut-off threshold. Calculate the weight growth rate. Predict the time when the shut-off threshold will be reached based on the weight growth rate. Activate the electromagnetic braking mechanism in advance to shut off the feeding port. Import the material into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, transmit the deviation between the actual feeding weight and the target weight to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

[0133] As described above, the computer program product provided in this application, through innovative design of a multi-feature fusion model, achieves accurate identification of material types via shape analysis and spectral detection. A material feeding control system is constructed, combined with multiple compensations, to establish a reliable weight monitoring mechanism. Adaptive learning is introduced, and through parameter updates and predictive shutdown, the accuracy of sample division is ensured. This method effectively solves the shortcomings of traditional technologies in material identification, material feeding control, and shutdown accuracy, providing technical support for dynamic sample division.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for precise control of dynamic sampling quantity for multiple varieties, characterized in that, The method includes: Grain images and near-infrared spectral data are acquired. Edge extraction and shape analysis are performed on the image data to extract the area, perimeter, and roundness features of the grain particles. Baseline correction and standardization are performed on the near-infrared spectral data. The shape features and spectral features are input into the variety identification model to identify the current material variety. Based on the material variety, the corresponding bulk density parameters, filling coefficient parameters, and shape influence coefficient parameters are retrieved from the parameter database. The parameters and the target feeding weight are input into the feeding control model. The target discharge volume is calculated based on the bulk density parameter. The target discharge volume is divided by the discharge volume per single turn of the rotary valve to obtain the integer number of turns. The remaining volume is calculated. The final opening angle is calculated based on the filling coefficient parameter and the shape influence coefficient parameter. The opening angle and the rotation speed parameter are input into the terminal control model to calculate the opening time. At the same time, multi-layer material level sensor data is collected. The current discharge weight is calculated based on the material level data and the bulk density parameter. The discharge weight is monitored in real time. The difference between the current feeding weight and the target feeding weight is input into the compensation model to calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. The compensation weight is subtracted from the target feeding weight to obtain the shut-off threshold. The weight growth rate is calculated, and the time when the shut-off threshold is reached is predicted based on the weight growth rate. The electromagnetic braking mechanism is activated in advance to shut off the feeding port, and the material is introduced into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, the deviation between the actual feeding weight and the target weight is transmitted to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

2. The method for precise control of dynamic sampling quantity for multiple varieties according to claim 1, characterized in that, The process involves acquiring grain images and near-infrared spectral data, performing edge extraction and shape analysis on the image data to extract the area, perimeter, and roundness features of the grain particles, and performing baseline correction and standardization on the near-infrared spectral data. The shape and spectral features are then input into a variety identification model to identify the current material variety, including: High-resolution images of grain samples were acquired under standard light source illumination. Gaussian filtering was applied to the images to eliminate noise. The Canny operator was used for edge detection to extract particle contours. The contours were converted into binary images. Morphological operations were performed on the binary images to remove edge burrs. Particle segmentation was performed based on connected component analysis. The pixel area, boundary perimeter, and roundness index of each segmented particle region were calculated as shape features. The light source of the near-infrared spectral acquisition device is turned on to scan the grain sample and obtain reflectance spectral data. The dark current signal is subtracted from the reflectance spectral data, whiteboard calibration is performed to eliminate instrument drift, and baseline offset is removed by multivariate scattering correction method. The corrected spectral data is resampled according to a preset wavelength interval, and the resampled data is subjected to standard normal transformation to generate standardized spectral features.

3. The method for precise control of dynamic sampling quantity for multiple varieties according to claim 1, characterized in that, The step of retrieving the corresponding bulk density parameter, filling factor parameter, and shape influence factor parameter from the parameter database according to the material type, and inputting the parameters and the target feeding weight into the feeding control model includes: The shape features of the grain sample are concatenated with the standardized spectral features to construct a feature vector. Principal component analysis is performed on the feature vector to reduce its dimensionality. The dimensionality-reduced features are then input into a pre-trained convolutional neural network to extract local feature maps. The feature maps are then input into a fully connected layer to calculate the probability score of each variety. The material variety identifier is determined based on the highest probability score. Based on the material type identifier, the corresponding record is retrieved from the parameter database, the basic bulk density value corresponding to the type is read, the basic bulk density is multiplied by the temperature and humidity correction coefficient to obtain the real-time bulk density parameter, the valve filling coefficient value and shape influence coefficient value corresponding to the type are read, and the real-time bulk density parameter, filling coefficient value and shape influence coefficient value are written into the parameter cache area of ​​the control model.

4. The method for precise control of dynamic sampling quantity for multiple varieties according to claim 1, characterized in that, The process of calculating the target discharge volume based on the bulk density parameter, dividing the target discharge volume by the discharge volume per single turn of the rotary valve to obtain an integer number of turns, calculating the remaining volume, and calculating the final opening angle based on the fill factor parameter and the shape influence factor parameter includes: The target feeding volume is calculated by the ratio of the target feeding weight to the real-time bulk density parameter. The target feeding volume is divided by the product of the number of valve cells and the volume of a single cell to obtain the theoretical feeding volume of a single valve turn. The actual feeding volume of a single valve turn is calculated by multiplying the theoretical feeding volume of a single valve turn by the filling coefficient parameter and the shape influence coefficient parameter. The target feeding volume is divided by the actual feeding volume of a single valve turn and rounded down to obtain the integer number of turns. The remaining volume is obtained by subtracting the product of the integer number of turns and the actual discharge volume per turn of the valve from the target discharge volume. The remaining volume is then divided by the volume of a single cell to calculate the number of cells to be opened. The number of cells to be opened is multiplied by the angle corresponding to a single cell to obtain the theoretical opening angle. Finally, the theoretical opening angle is calculated by multiplying it by the shape influence coefficient parameter.

5. The method for precise control of dynamic sampling quantity for multiple varieties according to claim 1, characterized in that, The process involves inputting the opening angle and rotation speed parameters into the terminal control model to calculate the opening time, simultaneously collecting data from multiple material level sensors, calculating the current material feeding weight based on the material level data and bulk density parameters, and monitoring the material feeding weight in real time, including: The basic rotation time is calculated by dividing the final opening angle by the rotation speed of the rotary valve. The sum of the basic rotation time and the material falling time is taken as the theoretical opening time. The actual opening time is obtained by adding the sum of the sensor response delay time and the control system delay time to the theoretical opening time. The operating parameters of the servo motor and the start-up sequence of the electromagnetic braking mechanism are set according to the actual opening time. Read the material level height data collected by the multi-layer material level sensor array, calculate the distance between each sensor position and the current material level height, assign sensor weight coefficients according to the distance, sum the product of the material level height data and the weight coefficients to obtain the weighted average material level height, and calculate the current material discharge weight by multiplying the weighted average material level height by the bottom area of ​​the receiving device, the real-time bulk density parameter, and the material surface distribution correction coefficient.

6. The method for precise control of dynamic sampling quantity for multiple varieties according to claim 1, characterized in that, The step involves inputting the difference between the current material loading weight and the target material loading weight into the compensation model to calculate the compensation weight for airborne materials, the system delay compensation weight, and the braking overshoot compensation weight. Subtracting the compensation weight from the target material loading weight yields the shut-off threshold, and the weight growth rate is calculated, including: The weight difference is obtained by subtracting the target weight from the current weight. A compensation coefficient is set based on the weight difference. The product of the valve opening area and the drop height is multiplied by the real-time bulk density parameter to obtain the air material compensation weight. The product of the material flow rate and the system response time is multiplied by the real-time bulk density parameter to obtain the system delay compensation weight. The product of the brake overshoot volume and the real-time bulk density parameter is obtained to obtain the brake overshoot compensation weight. The air material compensation weight, the system delay compensation weight, and the brake overshoot compensation weight are added together to obtain the total compensation weight. Collect the material feeding weight data of two adjacent sampling periods, calculate the difference between the material feeding weight data and divide it by the sampling time interval to obtain the weight growth rate, subtract the total compensation weight from the target material feeding weight to obtain the shutdown threshold, multiply the shutdown threshold by the system safety factor and subtract the minimum material feeding weight threshold to obtain the actual shutdown threshold, and calculate the time required for the current material feeding weight to reach the actual shutdown threshold based on the weight growth rate.

7. The method for precise control of dynamic sampling quantity for multiple varieties according to claim 1, characterized in that, The electromagnetic braking mechanism is activated in advance to shut off the feed port based on the predicted time of reaching the shut-off threshold according to the weight growth rate, and the material is introduced into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, the deviation between the actual feed weight and the target weight is transmitted to the adaptive learning model to update the filling coefficient parameter and shape influence coefficient parameter, including: The predicted weight is obtained by multiplying the weight growth rate by the time required to reach the shut-off threshold and adding the current feeding weight. The early shut-off time is calculated based on the predicted weight. The electromagnetic braking time, the control system delay time, and the material falling time are subtracted from the early shut-off time to obtain the actual shut-off trigger time. At the actual shut-off trigger time, the servo motor is stopped and the electromagnetic brake is activated to shut off the feeding port. The feeding distributor is driven to transport the feeding material to the bulk density measuring container, the moisture measuring container, and the imperfect particle measuring container respectively. The weight deviation is obtained by subtracting the target weight from the actual weight of the material. The relative error is obtained by calculating the ratio of the weight deviation to the target weight. The filling coefficient correction and shape influence coefficient correction are calculated based on the relative error. The filling coefficient correction is subtracted from the original filling coefficient parameter to obtain the updated filling coefficient parameter. The shape influence coefficient correction is subtracted from the original shape influence coefficient parameter to obtain the updated shape influence coefficient parameter.

8. A device for precise control of dynamic sample dispensing quantity for multiple varieties, characterized in that, The device includes: The feeding model determination module is used to collect grain images and near-infrared spectral data, perform edge extraction and shape analysis on the image data, extract the area, perimeter, and roundness features of the grain particles, perform baseline correction and standardization processing on the near-infrared spectral data, input the shape features and spectral features into the variety identification model, identify the current material variety, retrieve the corresponding bulk density parameters, filling coefficient parameters, and shape influence coefficient parameters from the parameter database according to the material variety, and input the parameters and the target feeding weight into the feeding control model. The feeding operation control module is used to calculate the target feeding volume based on the bulk density parameter, divide the target feeding volume by the feeding volume per turn of the rotary valve to obtain the integer number of turns, calculate the remaining volume, calculate the final opening angle according to the filling coefficient parameter and the shape influence coefficient parameter, input the opening angle and the rotation speed parameter into the terminal control model to calculate the opening time, and simultaneously collect data from multi-layer material level sensors, calculate the current feeding weight according to the material level data and the bulk density parameter, and monitor the feeding weight in real time. The adaptive shutdown module is used to input the difference between the current feeding weight and the target feeding weight into the compensation model, calculate the compensation weight for airborne material, the compensation weight for system delay, and the compensation weight for braking overshoot. The shutdown threshold is obtained by subtracting the compensation weight from the target feeding weight. The weight growth rate is calculated, and the time when the shutdown threshold is reached is predicted based on the weight growth rate. The electromagnetic braking mechanism is activated in advance to shut off the feeding port, and the material is introduced into the corresponding bulk density measuring container, moisture measuring container, and imperfect particle measuring container. Simultaneously, the deviation between the actual feeding weight and the target weight is transmitted to the adaptive learning model to update the filling coefficient parameter and the shape influence coefficient parameter.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method for precise control of dynamic sample quantity of multiple varieties as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for precise control of dynamic sample quantity of multiple varieties as described in any one of claims 1 to 7.

Images