A multi-cell astragalus detection and sorting method based on deep learning

Abstract

The application discloses a kind of based on deep learning's multi-cell astragalus detection sorting method and device, by prepositioning vibration blanking and dispersing mechanism, it is monomerized to peel off to pile astragalus, realizes regular loading of one slot one thing in the independent cell of introduction conveying belt;Control conveying belt step-by-step circulation operation, and collect still astragalus image, complete quality classification by pre-training deep learning model and bind cell logic number;Through Y type sorting insertion rod and cell guide rail mechanical coupling, replace bottom support, according to classification instruction trigger cell bottom hinged movable plate opening and closing, realize astragalus automatic blanking sorting.The present application takes PLC as bottom control core, asynchronously coordinates feeding, conveying, detection and sorting action, with abnormal monitoring and protective shutdown function.From physically eliminate astragalus stacking shielding and motion blur interference, significantly improve detection precision, realize detection and sorting integration closed loop automation, adapt astragalus large-scale intelligent processing, effectively improve sorting efficiency and quality stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent technology detection of Chinese medicinal materials. Specifically, it is a method that uses a multi-compartment structure to load Astragalus membranaceus, and uses a deep learning model to detect and sort the materials. The opening and closing of the bottom of the compartments is controlled by the results of the deep learning model inference, so as to achieve the effect of intelligent detection and automatic sorting. Background Technology

[0002] Astragalus is a commonly used traditional Chinese medicine, and its sorting process is one of the key steps to ensure its quality. In traditional astragalus sorting operations, manual sorting is often used to classify and distinguish astragalus based on its size, appearance, quality, and other indicators. This method is not only inefficient and difficult to keep up with the pace of large-scale production, but also prone to inconsistent sorting standards due to differences in human subjective judgment, which affects the quality stability of the traditional Chinese medicine.

[0003] With the development of intelligent detection technology, image-based detection methods are gradually being applied to the sorting of traditional Chinese medicine materials. However, existing solutions mostly directly detect batches of Astragalus membranaceus, which is prone to insufficient detection accuracy due to problems such as stacking obstruction and disordered arrangement of the Astragalus membranaceus. Furthermore, there is a lack of direct linkage between detection and sorting operations, requiring additional transfer and sorting equipment, which increases process complexity and cost. As a result, Astragalus membranaceus sorting remains difficult, lacks practicality, and is even more difficult to scale up for industrial enterprises or industries. For example, there are technologies in the published literature that use a conveyor belt grid coordinate system for defect location numbering (such as CN202512009786.4), but they do not achieve physical isolation loading of "one item per trough" or use step-by-step pause detection; there is also a vibration alignment feeding technology (such as CN202322254295.2), but it does not form a linkage with deep learning detection and grid bottom opening and closing sorting. The limitations of the above-mentioned solutions indicate that the existing technology has not yet solved the problems of insufficient detection accuracy caused by the stacking of Astragalus-related materials and motion ambiguity, as well as the low efficiency and losses caused by the separation of detection and sorting processes. Summary of the Invention

[0004] To address the limitations of traditional manual sorting, such as low efficiency, and insufficient linkage in existing intelligent detection solutions during the sorting process, this invention proposes a deep learning-based multi-compartment Astragalus detection and sorting method. This method achieves discretized processing of piled Astragalus and standardized loading of "one item per compartment" through the coordinated design of multi-compartment structure and deep learning detection algorithm. This effectively solves the problems of low detection accuracy caused by Astragalus stacking and disordered arrangement. At the same time, the automatic opening and closing design at the bottom of the compartments completes precise sorting, significantly improving the efficiency and accuracy of automated processing of Chinese medicinal materials.

[0005] The technical solution to achieve the objective of this invention is: A deep learning-based multi-compartment Astragalus detection and sorting method includes a multi-compartment Astragalus detection and sorting device, the method comprising the following steps: (1) Multi-compartment decentralized loading: Using a pre-vibration feeding and dispersing mechanism, the pile of Astragalus membranaceus is separated into individual pieces and guided into independent compartments integrated in the conveyor belt to achieve orderly loading of "one item per compartment". (2) Step-by-step steady-state detection: The control conveyor belt runs according to the step-by-step cycle logic of "run-stop-inspection". During the stop interval, the detection module is used to collect images and classify the quality of Astragalus membranaceus in the grid in a static state, and the classification results are associated with the logical number of the corresponding grid. (3) Mechanically coupled automated sorting: At the sorting station, the bottom support is replaced by the mechanical coupling of the separate sorting rod and the grid guide rail, and the opening and closing mechanism at the bottom of the grid is triggered according to the classification instruction to complete the automated unloading and sorting of Astragalus membranaceus.

[0006] Furthermore, it also includes system linkage control steps: (1) PLC is used as the underlying control core, and the feeding frequency, conveyor belt stepping time, camera triggering time and sorting execution action are coordinated through asynchronous communication mechanism; (2) The system monitors the status feedback of each actuator in real time. If an abnormality occurs, it will immediately trigger a protective shutdown, record the logical number of the faulty workstation, and issue an audible and visual alarm.

[0007] Going further: The bottom of the grid in step (1) adopts a downward-recessed opening and closing plate design, which uses gravity to guide and ensure that the Astragalus material is automatically in the center of the grid during the stepping and stopping of the conveyor belt; a linear guide rail for inserting rods is provided on one side of the bottom of the grid, which is physically connected to the inserting rod of the actuator; the bottom of the grid adopts a hinged movable plate structure, and is equipped with an electric control drive component to control the opening and closing of the movable plate.

[0008] The electronically controlled drive component is at least one of a stepper motor, an electromagnetic push rod, or a pneumatic component.

[0009] The method for image acquisition and quality classification of Astragalus membranaceus in a stationary state in step (2) is as follows: an industrial camera and an auxiliary light source are used to take pictures during the step-by-step pause. The images are input into a pre-trained deep learning target detection model to extract the morphology, color and surface features of Astragalus membranaceus and output sorting category instructions, which are then bound to the logical number of the current grid in real time.

[0010] The mechanically coupled automatic sorting method described in step (3) includes: 1) Non-working area support: In non-inspection / sorting areas, continuous support trays are installed under the conveyor belt to ensure that the movable plate of the grid is in a closed and locked state through physical support; 2) Work area coupling switching: At the sorting station, the actuator drives the "Y"-shaped sorting rod to move along a specific dimension and embed into the grid guide rail, taking over the support plate to bear the weight of the material, and serving as the physical fulcrum for the downward opening and closing of the movable plate.

[0011] The multi-compartment astragalus detection and sorting device includes: Stepper conveyor mechanism: includes a conveyor belt with multiple independent slots integrated, and a stepper drive motor for driving the intermittent operation of the conveyor belt; Image acquisition and processing unit: including industrial camera, auxiliary light source and controller equipped with deep learning algorithm model; Sorting execution mechanism: includes a continuous support tray set below the conveyor belt, and a mechanical coupling assembly set at the sorting station with a "Y"-shaped insert.

[0012] The image acquisition and processing unit also includes an adjustable bracket for adjusting the shooting distance and angle between the industrial camera and the grid surface. The auxiliary light source is an array of LED light sources with a constant color temperature.

[0013] The surface of the continuous support pallet is covered with a friction-reducing coating or a friction-reducing liner to reduce frictional loss between the movable plate of the grid and the pallet during the operation of the conveyor belt.

[0014] Through the above structure and method, the present invention has the following beneficial effects: 1. Achieving standardization of data sources through a breakthrough improvement in detection accuracy. By employing a physically isolated loading method of "multi-compartment + one item per compartment," problems such as material stacking, shading, and disordered arrangement common in traditional methods are physically eliminated. With the vibration mechanism and compartment dimensions properly coordinated, each piece of Astragalus membranaceus can independently enter the inspection station. Furthermore, the step-by-step "pause-inspection" logic eliminates motion blur, resulting in images with high signal-to-noise ratio and complete features obtained through deep learning models, ensuring high accuracy in identifying details such as texture, defects, and slice uniformity.

[0015] 2. Achieve full-process automation closed-loop by coupling mechanical processes and algorithms. From stacking multiple Astragalus membranaceus plants together to placing individual plants, to stably acquiring images of the Astragalus membranaceus areas, to real-time training and inference of the deep learning model, and finally to high-precision sorting of the Astragalus membranaceus, the entire Astragalus membranaceus defect detection and sorting process is a closed-loop linkage. The signals or information of the deep learning results of Astragalus membranaceus detection and sorting are transmitted to the PLC control system, dynamically synchronizing the stepping frequency of the conveyor belt and the model inference time. As soon as the Astragalus membranaceus detection command is issued after deep learning, the Astragalus membranaceus sorting execution mechanism responds promptly, achieving real-time response or extremely low latency to meet the production requirements of industrial enterprises. Through the integrated intelligent Astragalus membranaceus detection and sorting device, the problems of material waste and logical breakpoints caused by the separation of detection and sorting in traditional sorting solutions are solved.

[0016] 3. By optimizing the model through deep learning, the integrated intelligent detection and sorting capabilities of Astragalus membranaceus are enhanced. Combined with the continuous increase in Astragalus membranaceus sample capacity and the integration of transfer learning, the "appearance grading" is upgraded to more complex scenarios such as "insect and pest identification" and "foreign object removal", providing a new approach for large-scale Astragalus membranaceus sorting.

[0017] 4. The parameters of the grid unit and dispersing mechanism can be modularly adjusted or quickly replaced according to different specifications of Astragalus membranaceus (such as diameter, length, and thickness); the use of a support plate to support the bottom opening and closing plate of the grid solves the problem of vibration fatigue during the Astragalus membranaceus conveying process or the structural damage caused by the weight of the traditional material, ensuring reliable closure throughout the process. This integrated device is relatively easy to expand and upgrade in terms of hardware and software, and can be used to replace manual sorting of Astragalus membranaceus, thereby improving the economic benefits of industrial enterprises engaged in Astragalus membranaceus detection and sorting. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the conveyor belt grid structure according to an embodiment of the present invention; Detailed Implementation

[0019] The specific implementation process of the present invention will be described in detail below with reference to the technical solution. The core implementation method of the present invention is as follows: The deep learning-based multi-compartment Astragalus membranaceus detection and sorting method proposed in this invention achieves full automation of the Astragalus membranaceus detection and sorting process through the coordinated operation of four core components: multi-compartment distributed loading, step-by-step deep learning detection, mechanically coupled compartment sorting, and asynchronous linkage control of the system. The specific implementation details of each component are as follows: 1. Implementation of multi-compartment distributed loading Equipment Deployment and Structural Adaptation: A high-frequency vibrating feeder and multi-stage dispersive guide chutes are configured at the feed end of the conveyor belt. The main body of the conveyor belt integrates several independent slots, and the internal dimensions of the slots are customized according to the typical geometric parameters (length, width, and thickness) of Astragalus membranaceus slices, such as... Figure 1As shown. The bottom of the grid uses a hinged, downward-recessed movable plate structure. Its curvature or slope is designed according to the average diameter of the Astragalus slices, and a guide rail for the sorting rod is provided on one side of the movable plate to match the sorting rod.

[0020] The loading process is carried out in an orderly manner: the piles of Astragalus membranaceus awaiting sorting are individually processed by a vibration dispersing device and then fall one by one into the moving grid through guide chutes. By adjusting the matching relationship between the vibration frequency and the conveyor belt stepping speed, orderly loading of "one item per grid" is ensured, completely eliminating the overlap and obstruction interference between materials.

[0021] 2. Implementation of Step-by-Step Deep Learning Detection Inspection station status switching: The conveyor belt drive motor adopts a stepper control mode, following a cycle of "run - pause - image acquisition". When the grid slot moves precisely to the inspection station and enters the "pause" state, the system triggers an image acquisition command.

[0022] High-fidelity data acquisition: An industrial-grade high-definition CMOS camera and a high color rendering LED array light source are deployed in the detection area. During the step-pause intervals, the acquisition device captures images of Astragalus membranaceus in a stationary state within the grid, effectively avoiding dynamic blur.

[0023] Deep learning model inference: The acquired images are transmitted to the host computer in real time, and the pre-trained improved target detection model (such as the YOLO series or the Transformer architecture detection model) is input to extract the morphological features and quality indicators of Astragalus membranaceus, calculate the classification results and bind them to the logical number of the current grid in real time.

[0024] 3. Implementation of grid opening and closing sorting (core mechanical interaction) Pallet support in non-inspection areas: In the non-inspection / sorting areas of the conveyor belt (i.e., the regular transport path), continuous support pallets are laid under the grids. These pallets are in direct contact with the movable plates at the bottom of the grids, physically supporting the movable plates to ensure they remain in a stable closed state during transport and preventing materials from falling.

[0025] The insertion rod coupling in the working area: When the grid moves to the sorting station with the conveyor belt, the corresponding separate 'Y'-shaped insertion rod, driven by the actuator, moves along a specific dimension (such as horizontal or vertical) and embeds into the bottom guide rail of the grid, achieving a smooth transfer of support power. Then, the insertion rod is driven by the electronically controlled drive component. At this time, the insertion rod takes over the support plate to bear the weight of the material and serves as the physical fulcrum for subsequent opening and closing actions.

[0026] The sorting process is implemented in a coordinated manner: the control unit triggers the stepper motor at the corresponding workstation based on the category instructions output by the deep learning module. Guided and supported by the insert rod, the movable plate at the bottom of the slot opens downwards, allowing the astragalus root to fall precisely into the matching collection bin. After the action is completed, the insert rod guides the movable plate to reset and close, and then the slot steps away from the insert rod, is supported again by the support plate, and cycles back to the feeding end.

[0027] 4. Implementation of asynchronous linkage control in the system Multi-dimensional collaborative control: This invention uses a PLC as the underlying control core and coordinates the feeding speed, stepping frequency, camera triggering, and the action logic of the sorting actuator through an asynchronous communication mechanism.

[0028] Anomaly Monitoring and Closed-Loop Feedback: The system monitors the status feedback of the actuators in real time. If anomalies such as "grid jamming, moving plate not resetting, or network communication timeout" occur, the control device will immediately execute a protective shutdown and trigger an audible and visual alarm. Simultaneously, the system records the logical number of the abnormal workstation, enabling resume operation after the fault is resolved, ensuring the continuity and safety of the production process.

Claims

1. A deep learning-based multi-compartment Astragalus detection and sorting method, characterized by: The method includes the following steps: A multi-compartment Astragalus detection and sorting device. (1) Multi-compartment decentralized loading: Using a pre-vibration feeding and dispersing mechanism, the pile of Astragalus membranaceus is separated into individual parts and guided into independent compartments integrated in the conveyor belt; (2) Step-by-step steady-state detection: The control conveyor belt runs according to the step-by-step cycle logic of "run-stop-inspection". During the stop interval, the detection module is used to collect images and classify the quality of Astragalus membranaceus in the grid in a static state, and the classification results are associated with the logical number of the corresponding grid. (3) Mechanically coupled automated sorting: At the sorting station, the bottom support is replaced by the mechanical coupling of the separate sorting rod and the grid guide rail, and the opening and closing mechanism at the bottom of the grid is triggered according to the classification instruction to complete the automated unloading and sorting of Astragalus membranaceus.

2. The deep learning-based multi-compartment Astragalus detection and sorting method according to claim 1, characterized in that: It also includes system linkage control steps: (1) PLC is used as the underlying control core, and the feeding frequency, conveyor belt stepping time, camera triggering time and sorting execution action are coordinated through asynchronous communication mechanism; (2) The system monitors the status feedback of each actuator in real time. If an abnormality occurs, it will immediately trigger a protective shutdown, record the logical number of the faulty workstation, and issue an audible and visual alarm.

3. According to the deep learning-based multi-compartment Astragalus detection and sorting method described in claim 1, in step (1), the bottom of the compartment adopts a downwardly concave opening and closing plate design, which uses gravity guidance to ensure that the Astragalus material is automatically in the center position of the compartment during the stepping and stopping of the conveyor belt; a linear guide rail for inserting rods is provided on one side of the bottom of the compartment, which is physically connected to the inserting rod of the actuator; the bottom of the compartment adopts a hinged movable plate structure, and an electronically controlled drive component is configured to control the opening and closing of the movable plate.

4. The deep learning-based multi-compartment Astragalus detection and sorting method according to claim 2, wherein the electronically controlled drive component is at least one of a stepper motor, an electromagnetic push rod, or a pneumatic component.

5. According to the deep learning-based multi-slot Astragalus detection and sorting method of claim 1, the method of using the detection module to perform image acquisition and quality classification of Astragalus in a static state in the slot in step (2) is as follows: an industrial camera and an auxiliary light source are used to take pictures during the step-by-step pause, the images are input into a pre-trained deep learning target detection model, the morphology, color and surface features of Astragalus are extracted, and the sorting category instruction is output and bound to the logical number of the current slot in real time.

6. The deep learning-based multi-compartment Astragalus detection and sorting method according to claim 1, wherein the mechanically coupled automatic sorting method in step (3) includes: 1) Non-working area support: In non-inspection / sorting areas, continuous support trays are installed under the conveyor belt to ensure that the movable plate of the grid is in a closed and locked state through physical support; 2) Work area coupling switching: At the sorting station, the actuator drives the "Y"-shaped sorting rod to move along a specific dimension and embed into the grid guide rail, taking over the support pallet to bear the weight of the material, and serving as the physical fulcrum for the downward opening and closing of the movable plate.

7. The Astragalus detection and sorting device based on a multi-compartment structure as described in any one of claims 1-6, characterized in that: include: Stepper conveyor mechanism: includes a conveyor belt with multiple independent slots integrated, and a stepper drive motor for driving the intermittent operation of the conveyor belt; Image acquisition and processing unit: including industrial camera, auxiliary light source and controller equipped with deep learning algorithm model; Sorting execution mechanism: includes a continuous support tray set below the conveyor belt, and a mechanical coupling assembly set at the sorting station with a "Y"-shaped insert.

8. The deep learning-based multi-compartment Astragalus detection and sorting device according to claim 7, characterized in that: The image acquisition and processing unit also includes an adjustable bracket for adjusting the shooting distance and angle between the industrial camera and the grid surface. The auxiliary light source is an array of LED light sources with a constant color temperature.

Images