A fritillaria thunbergii non-destructive directional delivery system and method based on multi-dimensional feature fusion

Abstract

The application provides a Zhejiang Fritillaria unicum nondestructive directional delivery system and method based on multi-dimensional feature fusion, and belongs to the field of automatic processing of traditional Chinese medicinal materials. The application aims to solve the problems that Zhejiang Fritillaria and other medicinal materials are easy to be damaged in automatic feeding and are difficult to be recognized due to surface dirt. The system comprises a feeding unit, a visual detection unit and a directional placement unit. The feeding speed is dynamically adjusted according to the material density. The visual detection adopts double-layer contour matching, and if the confidence is lower than the threshold, micro-reversing secondary detection is triggered. The flexible gripper monitors the clamping force in real time, and once the limit is exceeded, the pressure is automatically released to protect the medicinal materials. Through the deep cooperation and closed loop of vision and force sense, the application realizes the efficient and nondestructive orientation of fragile and irregular medicinal materials, and greatly improves the material utilization rate and production efficiency.

Description

Technical Field

[0001] This application relates to the field of primary processing and automated equipment technology for Chinese medicinal materials, and in particular to a non-destructive directional delivery system and method for Fritillaria thunbergii based on multi-dimensional feature fusion. Background Technology

[0002] Fritillaria thunbergii is a commonly used traditional Chinese medicine for clearing heat and resolving phlegm. Its bulbs are typically flat, 5-8 mm thick and 15-35 mm in diameter, and often have soil residue or natural irregular textures on the surface. In the processing of traditional Chinese medicine slices, orientation and loading before slicing are key steps.

[0003] Currently, there are two main methods for pre-processing Zhejiang fritillary bulb: The first is manual feeding, where workers must manually identify the front and back sides and adjust the orientation, which is labor-intensive and inefficient, making it difficult to meet the needs of large-scale production. The second method uses general-purpose automated feeding equipment (such as vibratory feeders or general-purpose robotic arms), but this has significant drawbacks for the specific material of Zhejiang fritillary bulb.

[0004] 1. High rate of physical damage: The epidermis of Fritillaria thunbergii is extremely easy to break (the damage threshold is only about 5N). The rigid gripping or constant force clamping of general equipment can easily cause the medicinal material to be pinched or broken, affecting the grade and selling price of the medicinal material.

[0005] 2. Poor robustness of visual recognition: The surface of Fritillaria thunbergii often has blurred features due to mud or reflection. General visual algorithms usually directly identify such materials as waste and discard them, resulting in a great waste of materials; or due to the blind spots in a single viewpoint caused by the flat features, the grasping posture is incorrect.

[0006] Therefore, there is an urgent need for an intelligent device that can integrate visual and force feedback, protect fragile medicinal materials, and solve the recognition problem by actively adjusting the posture. Summary of the Invention

[0007] This application provides a non-destructive directional delivery system and method for Fritillaria thunbergii based on multi-dimensional feature fusion. It addresses the problems of existing technologies that mainly rely on inefficient manual operation, while general automated equipment lacks adaptive protection and fault tolerance mechanisms for fragile, flat, and surface-dirty medicinal materials, resulting in high breakage rates and persistently high material rejection rates due to recognition interference.

[0008] The core technology of this invention is to construct a closed-loop system of "visual-mechanical depth collaboration", which solves the problem of identification of complex surfaces through "double-layer contour matching" and "confidence-driven micro-flip secondary detection", and uses the "pressure threshold automatic pressure relief" mechanism to achieve non-destructive grasping of fragile medicinal materials.

[0009] Firstly, this application provides a non-destructive directional delivery system for Fritillaria thunbergii based on multi-dimensional feature fusion, comprising: The feeding unit is used to transport the medicinal materials to be processed. The visual inspection unit is located downstream of the feeding unit. It is used to collect image information of the medicinal materials and output the posture data and posture recognition confidence of the medicinal materials based on the image processing algorithm. The orientation placement unit includes a robotic arm and a flexible gripping mechanism located at the end of the robotic arm. The flexible gripping mechanism integrates a pressure detection module for real-time feedback of the gripping force. The main control unit is communicatively connected to the feeding unit, the vision inspection unit, and the orientation placement unit, respectively. The master control unit is configured to execute the following collaborative control logic: When the confidence level of posture recognition is lower than the preset validity threshold, the orientation placement unit is controlled to perform a micro-flipping action to change the posture of the medicinal material, and the vision detection unit is triggered to perform secondary image acquisition and recognition. When the clamping force reported by the pressure detection module exceeds the preset safety pressure threshold, the flexible clamping mechanism is controlled to perform an active pressure relief operation until the clamping force is reduced to a safe range.

[0010] Furthermore, the visual detection unit employs a two-layer contour matching algorithm to calculate pose data. The two-layer contour matching algorithm includes: Coarse matching layer: Based on the geometric features of the medicinal material outline, sector sampling and principal component analysis are performed to determine the principal axis direction and approximate orientation of the medicinal material; Fine matching layer: Based on coarse matching, the texture direction features and specific anatomical structure features of the surface of the medicinal material are extracted, and the front and back sides and rotation angle of the medicinal material are determined through multi-scale texture analysis. The confidence score for pose recognition is calculated by weighted fusion of the feature matching scores of the coarse matching layer and the fine matching layer.

[0011] Furthermore, specific anatomical features include the umbilicus depression region of the medicinal material; the visual detection unit includes two sets of camera components arranged symmetrically at a preset angle, and an adjustable brightness ring diffuse reflection light source surrounding the camera components; the two sets of camera components are used to capture top view and side view images of the medicinal material, respectively, and a three-dimensional posture model of the medicinal material is established through stereo vision fusion calculation.

[0012] Furthermore, the specific execution logic of the micro-flipping action is as follows: The robotic arm is controlled to drive the flexible gripping mechanism to grip the medicinal materials with a preset low gripping force lower than the normal gripping force; the end of the robotic arm is controlled to perform a spatial flipping motion with an angle of less than or equal to 30 degrees; during the flipping process, the gripping opening and closing degree is adjusted in real time according to the feedback of the pressure detection module to prevent the medicinal materials from slipping.

[0013] Furthermore, the flexible clamping mechanism includes three circumferentially distributed silicone fingers, the inner surface of which is provided with anti-slip micro-textures; the active pressure relief operation specifically includes: When the clamping force exceeds 8N, the pneumatic proportional control valve is controlled to discharge 50% of the air pressure in the air path, causing the clamping force to decrease instantly.

[0014] Furthermore, the feeding unit includes a roller feeding mechanism and a conveyor track connected thereto; The visual inspection unit is also configured to count the number of medicinal materials passing through the inspection area per unit time in real time to generate density data; The main control unit dynamically adjusts the conveying speed of the feeding unit based on density data: when the density data is higher than the preset dense threshold, the conveying speed is reduced; when the density data is lower than the preset sparse threshold, the conveying speed is increased.

[0015] Furthermore, it also includes a sample increment learning module, which is used to record the correct pose image data after manual intervention and use the image data as training samples to dynamically update the feature template library of the visual detection unit.

[0016] Furthermore, the silicone finger hardness of the flexible clamping mechanism is Shore 30-40, and the width of the conveying track is customized to match the diameter range of Fritillaria thunbergii, which is 35mm.

[0017] Secondly, this application provides a non-destructive directional delivery method for Fritillaria thunbergii based on multi-dimensional feature fusion, comprising the following steps: S1: The medicinal materials are transported to the visual inspection area via the feeding unit; S2: Use a visual detection unit to acquire images of medicinal materials and calculate the confidence level of pose recognition; S3: Determine whether the confidence level is greater than or equal to the preset validity threshold; if yes, proceed to step S5; if no, control the directional placement unit to grab the medicinal material and perform a micro-flip, then return to step S2 for a second detection. S4: If the confidence level of the second test is still lower than the validity threshold, an alarm signal will be issued to prompt manual intervention; S5: The main control unit plans the motion path based on the final confirmed posture data and controls the orientation placement unit to grab the medicinal materials; S6: During the gripping and handling process, the clamping force is monitored in real time. If the force exceeds the safe pressure threshold, an active pressure relief operation is immediately executed. S7: Place the medicinal materials in the target position according to the prescribed posture and send a feedback signal to trigger the next round of feeding.

[0018] Furthermore, the step of calculating the pose recognition confidence in step S2 specifically includes: Adaptive median filtering and local brightness equalization are applied to the acquired images; Extract the contour curvature features of the medicinal materials in the image and perform coarse matching with a standard ellipse template; Extract the scale texture and navel features of the medicinal materials in the image and perform fine matching with the pre-stored texture feature library; The overall confidence level is calculated by weighting the similarity scores of coarse matching and fine matching.

[0019] The main contributions and innovations of this invention are as follows: 1. Active fault tolerance based on "confidence-driven" significantly improves material utilization: Unlike the crude "reject if unclear" approach in existing technologies, this invention designs a hierarchical processing logic based on recognition confidence. When the recognition confidence of the medicinal material falls below 80% due to mud or occlusion on the surface, the system does not directly reject it. Instead, it triggers a robotic arm to perform a micro-rotation of ≤30° with a low gripping force (2-3N), coordinating with a camera for secondary data acquisition. This biomimetic logic of "using motion to aid vision" effectively solves the blind spots of static vision, significantly reducing the occurrence of good products being misjudged as waste.

[0020] 2. A unique "overload relief" force control mechanism achieves zero damage to fragile medicinal materials: Addressing the biological characteristic of Fritillaria thunbergii's easily damaged skin (≤5N), this invention integrates millisecond-level response force control protection into the flexible gripper. A built-in thin-film pressure sensor monitors the clamping force in real time. Once the force exceeds the safety threshold (8N), the controller immediately instructs the proportional regulating valve to automatically release 50% of the pressure, instantly restoring the clamping force to a safe range. This active compliant control completely solves the problem of mechanical damage to medicinal materials caused by the "one-size-fits-all" clamping of general-purpose equipment.

[0021] 3. Multi-dimensional visual feature fusion to overcome the challenge of complex surface recognition: Addressing the flat and complex texture of Fritillaria thunbergii, this invention employs a two-layer algorithm: "coarse matching (flattened round outline) + fine matching (umbilicus depression / scale texture)," combined with dual-camera 45° stereo imaging and a ring-shaped diffuse light source. This design effectively suppresses specular reflection interference from damp surfaces and ignores local mud spots and imperfections, accurately extracting key biological features such as the "umbilicus" and "scale orientation," significantly improving recognition accuracy under non-ideal conditions.

[0022] 4. Adaptive and Coordinated Feeding-Detection Process: This invention establishes a density-based dynamic speed control closed loop. The vision module calculates the material density and spacing on the conveyor belt in real time, directly feeding back to control the conveying speed of the feeding motor. It automatically slows down when material accumulation is detected and accelerates when materials are sparse, ensuring each piece of Zhejiang fritillary bulb has a sufficient processing time window (≥0.5s), avoiding the common problems of material blockage or idling in traditional equipment, and maximizing production efficiency.

[0023] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a structural diagram of the non-destructive directional delivery system for Fritillaria thunbergii based on multi-dimensional feature fusion according to an embodiment of this application; Figure 2 This is a structural diagram of a robotic arm according to an embodiment of this application; Figure 3 This is a diagram of a flexible gripper structure according to an embodiment of this application.

[0025] In the diagram, 1. Roller; 2. Vision camera; 3. Mother-of-pearl conveyor line; 4. Robotic arm; 5. Flexible gripper; 4-1-Robotic arm platform; 4-2-Motor; 4-3-Large arm body; 4-4-Small arm body; 4-5-Industrial camera; 5-1-Mounting flange; 5-2-Drive cylinder; 5-3-Telescopic push rod; 5-4-Fixed support; 5-5-First stage connecting rod; 5-6-Connecting rod gasket; 5-7-Connecting rod pin; 5-8-Connecting rod fastening screw; 5-9-Gripper body; 5-10-Miniature thin-film pressure sensor; 5-11-Gripper contact pad. Detailed Implementation

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0027] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0028] Example 1: A non-destructive directional delivery system for Fritillaria thunbergii based on multi-dimensional feature fusion This embodiment provides an intelligent directional delivery system for flat rhizomes (taking Fritillaria thunbergii as an example). The system mainly consists of a feeding unit, a vision detection unit, a directional placement unit, and a main control unit (industrial computer). Each unit is connected via industrial Ethernet or bus communication to form a deep collaborative closed loop of "feeding-detection-direction".

[0029] 1. Feeding Unit: The feeding unit is responsible for the orderly conveying of materials. For example... Figure 1 As shown, it mainly includes a roller 1 and a mother-of-pearl conveyor line 3 connected to it.

[0030] Roller 1 (roller feeding mechanism): Roller 1 is made of 304 stainless steel with a sandblasted surface (roughness Ra=1.6-3.2μm), which increases the friction between the roller and the medicinal materials to facilitate lifting, while avoiding the rigid wear of the medicinal material surface caused by the smooth stainless steel. The tilt angle of roller 1 is set to 3-5°.

[0031] Fritillaria cirrhosa conveyor line 3 (conveyor track): connected to the discharge port of roller 1. In order to adapt to the flat characteristics of Fritillaria thunbergii (diameter 15-35mm, thickness 5-8mm), the track width is customized to 35mm, and there are 10mm high guide plates on both sides, which force the medicinal materials to pass through in a single row in a "flat" posture to prevent stacking.

[0032] Density-based speed control mechanism: The drive motor of the fritillaria conveyor line 3 is connected to a frequency converter. The main control unit dynamically adjusts the speed of the fritillaria conveyor line 3 using a PID algorithm based on the "material density data" (i.e., the amount of material passing through the detection zone per unit time) fed back by the downstream visual inspection unit. For example, the initial speed is set to 0.5 m / s; when the detected material density exceeds a preset density threshold (e.g., ≥2 pieces / second), the speed is automatically reduced by 20% to 0.4 m / s; when the material density is sparse (<1 piece / second), the speed is increased to 0.6 m / s. This ensures that each piece of medicinal material has at least a 0.5-second dwell / processing window in the detection zone.

[0033] 2. Visual inspection unit The visual inspection unit is set in the temporary storage area at the end of the fritillaria conveyor line 3 to acquire the three-dimensional posture information of the medicinal material.

[0034] Hardware configuration: Primarily includes a fixed vision camera 2. To address the visual blind spots that are prone to occur with flat medicinal materials, this embodiment employs a dual-camera stereo vision solution. Two sets of vision cameras 2 are symmetrically arranged at a 45° angle above the temporary storage area, capturing the top view and oblique side view of the medicinal materials, respectively. Furthermore, during subsequent fine processing or micro-flipping processes, industrial cameras 4-5 mounted on the robotic arm can be used for close-range observation.

[0035] Light source system: To address the issue of mirror reflection caused by the water content (8-12%) on the surface of Fritillaria thunbergii, a ring-shaped LED light source with multi-level adjustable brightness is configured around the lens of the vision camera 2, and a diffuser is added to generate diffused light, so that the light illuminating the surface of the medicinal material is uniform and the bright spots are suppressed.

[0036] Background design: The temporary storage area uses a matte dark material as the base plate, which contrasts with the gray and white Fritillaria thunbergii to facilitate contour extraction.

[0037] The temporary storage area is located directly below the vision camera 2, in the section of the fritillary conveyor line 3 that connects the discharge port of the roller 1 to the slicing worktable.

[0038] 3. Oriented placement unit This unit is responsible for grasping, adjusting the posture of the medicinal materials and placing them in a directional manner. Its core components are the robotic arm 4 and the flexible gripper 5 (flexible gripping mechanism) mounted at its end.

[0039] Robotic arm 4 structure: such as Figure 2 As shown, the robotic arm 4 includes a robotic arm platform 4-1, with a motor 4-2 inside or on the side serving as a power source. A main arm body 4-3 is movably connected to the robotic arm platform 4-1, and a forearm body 4-4 is connected to the front end of the main arm body 4-3, forming a multi-degree-of-freedom motion mechanism. At the end of the forearm body 4-4, an industrial camera 4-5 and a flexible gripper 5 serving as an end effector are integrated.

[0040] Detailed structure of flexible gripper 5 (flexible clamping mechanism): as follows Figure 3 As shown, in order to achieve non-destructive gripping of fragile Zhejiang fritillary bulbs, the flexible gripper 5 adopts a precise transmission and buffer design.

[0041] Connection and Drive: The flexible gripper 5 is mounted to the end of the robotic arm 4 via a mounting flange 5-1. Its power comes from the drive cylinder 5-2, which drives the telescopic push rod 5-3 to perform linear reciprocating motion.

[0042] Transmission Linkage: The movement of the telescopic push rod 5-3 is transmitted through a linkage mechanism, which includes a fixed support 5-4 and a primary linkage 5-5. The components are rotatably connected by linkage connecting pins 5-7, and are equipped with linkage connecting washers 5-6 to reduce wear. They are locked in place by linkage fastening screws 5-8 to ensure transmission accuracy.

[0043] Clamping assembly: A linkage mechanism drives the opening and closing of the gripper body 5-9 (i.e., the three-lobed base). At the part in contact with the medicinal material, there is a soft gripper contact pad 5-11 (such as a silicone finger, Shore hardness 30-40), with anti-slip micro-textures on the surface.

[0044] Tactile feedback (force control core): A miniature thin-film pressure sensor 5-10 (pressure detection module) is integrated below or inside the gripper contact pad 5-11. This sensor can collect the gripping force in real time and feed it back to the main control unit. When the force exceeds the 8N threshold, the control system adjusts the air pressure of the drive cylinder 5-2 to actively release pressure.

[0045] Example 2: Visual Recognition Algorithm Based on Two-Layer Matching This embodiment details the visual recognition algorithm running in the main control unit, which aims to solve the recognition difficulties caused by the irregular surface and mud adhering to Fritillaria thunbergii. This embodiment describes in detail the visual recognition algorithm running in the main control unit using visual camera 2 (or assisted by industrial cameras 4-5). The algorithm flow is as follows: 1. Image preprocessing: Adaptive median filtering is performed on the acquired image (based on vision camera 2) to remove mud noise; then local brightness equalization is performed to eliminate shadow interference.

[0046] 2. Two-layer contour matching (core algorithm): First layer: Coarse matching. Based on geometric shape features, the algorithm extracts the outer edge contour of the medicinal material and fits the minimum bounding ellipse. Principal component analysis (PCA) is used to calculate the direction of the first principal component, determining the major axis direction (approximate orientation) of the medicinal material. This step is insensitive to local incompleteness of the contour.

[0047] For example, we can prioritize extracting approximate elliptical or oval contours from the image and estimate their approximate orientation using sector sampling and principal component analysis (PCA). This stage allows for local incompleteness in the contours; only the overall curvature needs to be matched to preliminarily determine the orientation. The sector sampling steps are as follows: The outline of Fritillaria thunbergii was extracted from the preprocessed image. The minimum circumcircle was fitted to obtain the center origin and radius R, and the effective sampling range was limited to 0.6R-1.0R. Twelve equal-angled sectors of 30° were drawn with the origin as the center. The number of feature points in the outline of each sector was counted and assigned a weight of 0-1.0 according to the point density. The 2-3 adjacent sectors with the highest weighted point density were selected, and their central polar angle was taken as the candidate direction. The feature points in the candidate sectors were input into PCA to extract the feature vector corresponding to the largest feature value, and the approximate orientation of the long axis of the flat end of Fritillaria thunbergii was output to provide a reference for fine matching.

[0048] The second layer: fine matching. Based on biological texture features. Within the region of interest (ROI) determined by coarse matching, the algorithm further searches for the unique "umbilicus depression" and "scale texture direction" of Fritillaria thunbergii. Specifically, a multi-scale texture analysis module (such as using a Gabor filter bank or a gray-level co-occurrence matrix GLCM) is used to extract the direction with the strongest texture energy as the growth direction of the scales, and the depression area with the most dramatic gray-level gradient change in the image is identified as the umbilicus location.

[0049] In this embodiment, the multi-scale texture analysis module is specifically as follows: Scale setting: Adapted to the size of Fritillaria thunbergii, three core scales are set (small scale 3×3, medium scale 7×7, large scale 11×11), and the scales are correlated through Gaussian pyramid downsampling.

[0050] Texture extraction: The Sobel operator is used to calculate the pixel gradient direction at each scale. The radial gradient distribution is statistically analyzed for the navel depression area (ring texture), and the tangential gradient is statistically analyzed for the scale area (parallel texture). The texture direction histograms at each scale are generated.

[0051] Consistency Verification: Cross-scale comparison of histogram peak directions. If the peak deviation across three scales is ≤5°, it is considered a valid texture; if the deviation is >5°, it is considered a defect, and the data in that area is automatically masked. Orientation Locking: Based on the consistent orientation of valid textures, combined with the major axis orientation of the coarse matching layer, the front and back sides and rotation angles are accurately output.

[0052] Confidence assessment: The system performs a weighted fusion of the shape similarity of the coarse match and the texture consistency of the fine match, and outputs a comprehensive confidence score of 0-100%.

[0053] If the confidence level is ≥80% (preset validity threshold), the pose recognition is considered successful, and the (x,y,z,angle) coordinates are output.

[0054] If the confidence level is less than 80% (usually due to mud obscuring the texture on the surface or the posture being too tilted), it is determined to be in a "fuzzy state" and triggers the micro-flipping process (see Example 3 for details). For example, the robotic arm 4 is triggered to drive the flexible gripper 5 to perform micro-flipping, and the vision camera 2 (or industrial camera 4-5) acquires the image a second time.

[0055] Incremental learning: The system runs an incremental learning module in the background. When manual intervention occurs in subsequent stages, the system automatically records the original image of the failed recognition and the correct label after manual correction, updates the feature template library, and improves the algorithm's recognition rate for irregularly shaped medicinal materials with increasing usage time.

[0056] In this embodiment, the internal feature template library is the core database storing the standard pose-feature mapping relationship of Fritillaria thunbergii, providing a comparison benchmark for coarse matching, fine matching, and stereo vision fusion. Its initial state is constructed based on 1000+ sets of standard Fritillaria thunbergii samples of different sizes and surface conditions, and the structure is divided into two layers: Coarse matching template layer: stores the elliptical / flattened circle contour curvature features of the flat end under standard pose; Fine-match template layer: Stores the contour parameters of the umbilicus depression area and the scale texture direction vector of the standard front and back sides. The template library supports dynamic addition and iterative optimization of feature vectors to adapt to individual differences of Fritillaria thunbergii and changes in detection scenarios.

[0057] The specific steps for dynamically updating the internal feature template library are as follows: Manual intervention in data collection: When the confidence level is still <80% after the second test, the posture of Fritillaria thunbergii is manually corrected to the standard state and confirmed. The system then automatically records three types of data for the sample: the original image, the corrected standard posture parameters, and the preprocessed contour and texture feature data. Validity screening: By filtering based on feature similarity thresholds, abnormal samples are removed, and only valid samples are retained; Feature extraction and standardization: For valid samples, the adaptive median filtering and local brightness equalization preprocessing described above are used to extract the contour curvature features of the coarse matching layer and the navel and scale texture features of the fine matching layer, which are then converted into feature vectors of a unified dimension. Template update strategy: Incremental addition: If the similarity between the feature vector of this sample and all existing templates in the library is less than 90%, it is directly appended to the corresponding category in the template library; Iterative optimization: If the similarity is ≥90%, calculate the mean of the feature vectors of the sample and the corresponding template, update the original template with the new mean, and improve the robustness of the template; Weight adjustment: The initial weight of newly added / updated templates is set to 0.8, and it gradually increases to 1.0 as the matching success rate increases in subsequent detections. The weight of old templates that have not been matched and have been used more than 1000 times is reduced to 0.5 to avoid redundancy. Real-time synchronization: After the template library is updated, it is immediately synchronized to the comparison algorithm of the visual inspection module. The new template can be directly used for the posture recognition of the next Fritillaria thunbergii, realizing dynamic improvement of adaptability.

[0058] Example 3: Control Flow of Intelligent Delivery Method This embodiment describes the control logic of the system in actual operation, especially the implementation process of "active fault tolerance" and "lossless force control".

[0059] Step S1: Collaborative feeding. Roller 1 works in conjunction with the fritillary bulb conveyor line 3 to transport the Zhejiang fritillary bulbs to the visual storage area. During this process, the speed is dynamically adjusted based on the density feedback from the visual system to prevent material accumulation.

[0060] Step S2: Initial Detection and Decision. The visual camera 2 (i.e., the high-definition camera) executes the algorithm described in Example 2.

[0061] If the confidence level is acceptable, robotic arm 4 will directly grasp the object.

[0062] Micro-flip logic (active fault tolerance): If the confidence level is less than 80%, the main control unit does not issue a rejection command, but instead instructs robotic arm 4 to move above the medicinal material.

[0063] Action 1: The grippers gently contact the medicinal material with a preset low gripping force (e.g., 2-3N).

[0064] Action 2: The robotic arm's end effector performs a small-angle flip (e.g., a 30° rotation around the horizontal axis), or picks up the medicinal material, rotates it in the air at a certain angle, and then puts it down. This process aims to expose the obscured features of the medicinal material or change the angle of reflection.

[0065] Action 3: After the flip is completed, the robotic arm returns to the safe position, triggering the camera to perform secondary image acquisition.

[0066] Action 4: Recalculate the confidence level. If the confidence level is acceptable at this point, continue with the subsequent process; if it is still unacceptable, discard it as waste or trigger an audible and visual alarm for manual handling.

[0067] Step S3: Attitude adjustment and non-destructive grasping.

[0068] Based on the final determined posture data, the robotic arm 4 adjusts the angle (rotation and translation) of the flexible gripper 5 so that it is perpendicular to the flat surface of the medicinal material for grasping.

[0069] Overload relief logic (non-destructive core): During the closing process of flexible gripper 5, the pressure sensor monitors in real time.

[0070] Under normal circumstances, the clamping force is controlled within the range of 2-8N.

[0071] Anomaly Handling: The miniature thin-film pressure sensor 5-10 monitors the pressure in real time. Once a sudden change in clamping force is detected and exceeds 8N (the safe pressure threshold) (e.g., the sharp edges of the medicinal material or foreign objects are clamped), the main control unit sends a command to the pneumatic proportional valve (drive cylinder 5-2) within milliseconds to perform an active pressure relief operation, that is, to instantly release 50% of the air pressure in the air circuit, so that the clamping force decreases rapidly and prevents the relatively soft skin of Fritillaria thunbergii from cracking.

[0072] Step S4: Oriented Placement. The robotic arm 4 transports the medicinal material to the slicing table and places it stably according to the orientation required by the slicer (e.g., the long axis perpendicular to the blade). Once placed in position, the sensor sends a signal, triggering the system to feed the next material in a closed-loop manner.

[0073] Example 4: Parameter Optimization Example For the specific material Fritillaria thunbergii, the optimal parameter combination of this system, verified through experiments, is as follows: Material characteristics: Zhejiang fritillary bulb has a thickness of 5-8mm, a diameter of 15-35mm, a skin breakage threshold of about 5N, and a moisture content of 8-12%.

[0074] Hardware parameters: Mother-of-pearl conveyor line 3 width: 35mm (±1mm).

[0075] Flexible gripper 5 hardness: Shore 35.

[0076] The spacing between the anti-slip textures on the gripper fingers is 0.5mm.

[0077] Control parameters: Feeding and testing cycle time matching: single-piece processing time ≥ 0.5s.

[0078] Micro-flip trigger threshold: confidence level < 80%.

[0079] Micro-flip angle: ≤30°.

[0080] Clamping force during micro-flipping: 2.5N.

[0081] Safety pressure threshold (overload protection): 8N.

[0082] Automatic pressure relief ratio: 50% (i.e., reduced to approximately 4N).

[0083] Those skilled in the art will understand that the above embodiments are merely preferred embodiments of the present invention. For other similarly shaped, flat rhizome medicinal materials (such as Corydalis yanhusuo and Atractylodes macrocephala), the system and method of the present invention can also be applied by simply adjusting the track width, gripper size, and corresponding force control threshold, and these adjustments should all be included within the scope of protection of the present invention.

[0084] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0085] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A non-destructive directional delivery system for Fritillaria thunbergii based on multi-dimensional feature fusion, characterized in that, include: The feeding unit is used to transport the medicinal materials to be processed. A visual inspection unit is located downstream of the feeding unit and is used to collect image information of the medicinal materials and output the posture data and posture recognition confidence of the medicinal materials based on the image processing algorithm. The directional placement unit includes a robotic arm and a flexible clamping mechanism located at the end of the robotic arm. The flexible clamping mechanism integrates a pressure detection module for real-time feedback of clamping force. The main control unit is communicatively connected to the feeding unit, the vision inspection unit, and the orientation placement unit, respectively. The main control unit is configured to execute the following collaborative control logic: When the confidence level of the posture recognition is lower than the preset validity threshold, the orientation placement unit is controlled to perform a micro-flipping action to change the posture of the medicinal material, and the visual detection unit is triggered to perform secondary image acquisition and recognition. When the clamping force reported by the pressure detection module exceeds the preset safety pressure threshold, the flexible clamping mechanism is controlled to perform an active pressure relief operation until the clamping force is reduced to a safe range. The visual detection unit uses a two-layer contour matching algorithm to calculate the pose data. The two-layer contour matching algorithm includes: Coarse matching layer: Based on the geometric features of the medicinal material outline, sector sampling and principal component analysis are performed to determine the principal axis direction and approximate orientation of the medicinal material; Fine matching layer: Based on coarse matching, the texture direction features and specific anatomical structure features of the surface of the medicinal material are extracted, and the front and back sides and rotation angle of the medicinal material are determined through multi-scale texture analysis. The confidence score for pose recognition is calculated by weighted fusion of the feature matching scores of the coarse matching layer and the fine matching layer.

2. The non-destructive directional delivery system for Fritillaria thunbergii as described in claim 1, characterized in that, The specific anatomical features include the umbilicus depression region of the medicinal material; the visual detection unit includes two sets of camera components arranged symmetrically at a preset angle, and an adjustable brightness ring diffuse reflection light source surrounding the camera components; the two sets of camera components are used to capture top view and side view images of the medicinal material, respectively, and a three-dimensional posture model of the medicinal material is established through stereoscopic vision fusion calculation.

3. The non-destructive directional delivery system for Fritillaria thunbergii as described in claim 1, characterized in that, The specific execution logic of the micro-flipping action is as follows: The robotic arm is controlled to drive the flexible gripping mechanism to grip the medicinal materials with a preset low gripping force lower than the normal gripping force; the end of the robotic arm is controlled to perform a spatial flipping motion with an angle less than or equal to 30 degrees; during the flipping process, the gripping opening and closing degree is adjusted in real time according to the feedback of the pressure detection module to prevent the medicinal materials from slipping.

4. The non-destructive directional delivery system for Fritillaria thunbergii as described in claim 1, characterized in that, The flexible clamping mechanism includes three circumferentially distributed silicone fingers, the inner surface of which is provided with anti-slip micro-textures; the active pressure relief operation specifically includes: When the clamping force exceeds 8N, the pneumatic proportional control valve is controlled to discharge 50% of the air pressure in the air path, causing the clamping force to decrease instantly.

5. The non-destructive directional delivery system for Fritillaria thunbergii as described in claim 4, characterized in that, The feeding unit includes a roller feeding mechanism and a conveying track connected thereto; The visual detection unit is also configured to count the number of medicinal materials passing through the detection area in real time to generate density data. The main control unit dynamically adjusts the conveying speed of the feeding unit based on the density data: when the density data is higher than the preset dense threshold, the conveying speed is reduced; when the density data is lower than the preset sparse threshold, the conveying speed is increased.

6. The non-destructive directional delivery system for Fritillaria thunbergii as described in claim 1, characterized in that, It also includes a sample incremental learning module, which is used to record the correct posture image data after manual intervention and use the image data as training samples to dynamically update the feature template library of the visual detection unit.

7. The non-destructive directional delivery system for Fritillaria thunbergii as described in claim 5, characterized in that, The silicone of the flexible clamping mechanism has a Shore hardness of 30-40, and the width of the conveying track is customized to match the diameter of the Fritillaria thunbergii, which is 35mm.

8. A delivery method using the non-destructive directional delivery system for Fritillaria thunbergii as described in any one of claims 1-7, characterized in that, Includes the following steps: S1: The medicinal materials are transported to the visual inspection area via the feeding unit; S2: Use a visual detection unit to acquire images of medicinal materials and calculate the confidence level of pose recognition; S3: Determine whether the confidence level is greater than or equal to the preset validity threshold; if yes, proceed to step S5; if no, control the directional placement unit to grab the medicinal material and perform a micro-flip, then return to step S2 for a second detection. S4: If the confidence level of the second test is still lower than the validity threshold, an alarm signal will be issued to prompt manual intervention; S5: The main control unit plans the motion path based on the final confirmed posture data and controls the orientation placement unit to grab the medicinal materials; S6: During the gripping and handling process, the clamping force is monitored in real time. If the force exceeds the safe pressure threshold, an active pressure relief operation is immediately executed. S7: Place the medicinal materials in the target position according to the prescribed posture and send a feedback signal to trigger the next round of feeding.

9. The delivery method according to claim 8, characterized in that, The step of calculating the pose recognition confidence level in step S2 specifically includes: Adaptive median filtering and local brightness equalization are applied to the acquired images; Extract the contour curvature features of the medicinal materials in the image and perform coarse matching with a standard ellipse template; Extract the scale texture and navel features of the medicinal materials in the image and perform fine matching with the pre-stored texture feature library; The overall confidence level is calculated by weighting the similarity scores of coarse matching and fine matching.

Images