A computer vision-based vertical pak choi purification processing and packaging all-in-one machine

Abstract

This invention discloses a vertical integrated machine for purifying, processing, and packaging bok choy based on computer vision, belonging to the field of intelligent equipment for fruit and vegetable processing. It solves the problems of high cost, low precision, easy spoilage, and insufficient automation in existing leafy vegetable pretreatment methods. The machine has a vertical four-layer integrated structure, including a motion execution system, a control system, and a vision perception system. The motion execution system integrates flexible grippers, gears and racks, a flap door, brushes, and a packaging mechanism. Through vertical gravity transport, it sequentially completes root cutting, brush cleaning, identification and removal of rotten and yellowed leaves, and unsealed air-blown bagging. The vision system uses a Raspberry Pi 5 equipped with a lightweight YOLOv8 model and a Picamera camera to achieve accurate identification of rotten and yellowed leaves. The control system has a built-in automatic startup program. This invention features a compact structure, low cost, and low energy consumption, with a single plant processing time of ≤10 seconds, making it suitable for small-scale fruit and vegetable processing workshops and farmers' markets.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment technology for fruit and vegetable processing, specifically a vertical integrated machine for purifying, processing, and packaging bok choy based on computer vision. It is suitable for the integrated processing of leafy vegetables such as bok choy, including root cutting, cleaning, removal of rotten and yellow leaves, and unsealed bagging and packaging. It falls under the category of cross-application of agricultural machinery and computer vision. Background Technology

[0002] As a common leafy vegetable, bok choy has a large market demand, and its pre-processing (root cutting, cleaning, removing rotten leaves, and packaging) is a key step in its commercial distribution. Currently, the pre-processing of bok choy mostly relies on manual operation or segmented processing using traditional equipment, which has many technical drawbacks: manual processing is inefficient, with a skilled worker only able to process up to 200 jin (100 catties) of bok choy per day, and the identification and removal of rotten and yellow leaves depends on experience, making it prone to missed or incorrect judgments; traditional equipment often uses conveyor belts for material transport, resulting in high overall equipment costs and energy consumption, and the cleaning process often uses bubble tank cleaning solutions, with a 1.5m long bubble tank costing 12,000 yuan, requiring additional cylinders and water circulation systems, and residual moisture on the surface of the vegetables after cleaning can accelerate spoilage and reduce shelf life; at the same time, the visual recognition process of existing bok choy pre-processing equipment often directly detects uncleaned vegetables, and dirt and dust are easily misjudged as rotten leaves or insect holes, resulting in low recognition accuracy; in addition, traditional packaging equipment often integrates sealing processes, which are complex in structure, and for bok choy sold in bulk, the sealing process is meaningless, only increasing equipment costs and processing time. To address the aforementioned issues, there is an urgent need to develop a low-cost, low-energy-consumption, and highly automated integrated pre-processing equipment for bok choy. This equipment would enable continuous operations including root cutting, cleaning, visual removal of rotten leaves, and packaging, while avoiding the technical shortcomings of traditional conveyor belts and air-filled tanks. It would also improve visual recognition accuracy and adapt to the unsealed packaging requirements for bulk sales. Each 0.4-pound bok choy head can be processed in just 10 seconds, and one machine operating on two shifts can process 2304 pounds of bok choy per day. Summary of the Invention

[0003] The purpose of this invention is to provide a vertical integrated machine for purifying, processing, and packaging bok choy based on computer vision, to solve the technical problems of low pre-processing efficiency, high equipment cost, low visual recognition accuracy, and easy spoilage after cleaning. This invention adopts a vertical four-layer integrated structure, using pipes to achieve gravity-feed transportation of bok choy, replacing the traditional conveyor belt structure and significantly reducing equipment cost and energy consumption. It abandons the bubble tank cleaning solution and uses food-grade brush rollers for cleaning, effectively removing impurities and weak, rotten yellow leaves while avoiding residual moisture that accelerates vegetable spoilage. The cleaning process is placed before visual recognition, eliminating interference from dirt and dust on recognition accuracy. A Raspberry Pi 5 with a lightweight YOLOv8 model and Picamera camera module is used for accurate identification of rotten yellow leaves, combined with flexible grippers to simulate manual tearing for leaf removal. The packaging process uses a blower-blown bag + gravity-push bag method without sealing, suitable for bulk sales. The entire process is automated, with a small footprint, compact structure, and no need for upper-level computer assistance, significantly improving the pre-processing efficiency of bok choy.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: like Figure 1 A vertical integrated machine for purifying, processing, and packaging bok choy based on computer vision is disclosed. It includes a motion execution system, a control system, and a visual perception and information processing system. The machine has a vertical four-layer integrated structure, suitable for processing bok choy with stems up to 5cm in diameter. The bok choy is transported by gravity through vertical pipes between layers. Along the material transport direction, the machine sequentially completes the following processes: root cutting, brush roller cleaning, visual identification and removal of rotten and yellow leaves, and gravity bagging without seals. The machine has a total height of 1450mm and a horizontal cross-sectional dimension of 910×455mm. It operates fully automatically without the need for a host computer.

[0005] I. Motion Execution System The motion execution system consists of a fuselage frame, gravity transmission pipes, root cutting mechanism, brush cleaning mechanism, rotten and yellow leaf removal mechanism, and blower packaging mechanism.

[0006] like Figure 2 The machine frame is constructed from 20×20×R1.0 industrial aluminum profiles. The frame has four vertical columns spaced 355×355mm apart. The four working areas have heights of 150mm, 573mm, and 369mm respectively. The aluminum profiles are secured to each other and to other components using both angle brackets and bolts / nuts, as well as aluminum profile corner groove connectors and screws. Figure 3 Corner slot connections are used to ensure a secure connection of the aluminum profiles, guaranteeing the strength of the frame connection and its reliability during operation. Each layer is equipped with a working mechanism and pipe installation base, with a base load-bearing capacity of ≥15kg.

[0007] like Figure 4The gravity transfer pipeline is a four-layer rigid food-grade PP vertical pipe with an internal cross-section of 5cm × 7cm and a wall thickness of 5mm. Each layer of the pipeline is arranged vertically and coaxially, with a flared transition structure on all four sides of the top hopper, with a transition opening diameter of 10cm. The bottom of the third-layer pipeline is a sloping pipe with an inclination angle of 30°, seamlessly connecting to the fourth-layer packaging system. All pipeline inner walls are smoothly treated, with a friction coefficient ≤0.2.

[0008] like Figure 5 The first-layer root-cutting mechanism includes a feeding hopper, flexible fixed grippers (MG996R servo motor driven), a gear and rack root-cutting assembly (SG90 servo motor driven), and an SG90 servo motor-controlled flap door. The feeding hopper is inverted conical with a 10×10cm top opening, seamlessly connecting to the top of the pipe. An 8×5cm installation port is provided on the back of the pipe, housing a flexible fixed gripper driven by an MG996R servo motor, with a clamping stroke of 1.5-72mm, a torque of 13kg·cm, and a lever arm length of 72mm at the maximum opening angle of the gripper. The gravitational acceleration g is taken as 9.8m / s². 2 Substituting into the formula, we get F≈17.7N. A 6×4cm root-cutting opening is made on the front of the pipe, 3.5cm from the bottom. A rack and pinion root-cutting assembly is installed on the outside: the rack is driven by an SG90 servo motor to reciprocate. The rack is 15cm long with an effective stroke of 9.7cm. A 3mm thick hard stainless steel root-cutting blade is fixed at the end with a 4mm bolt. The blade sharpness is ≥HRC55. A 4.5cm diameter square flap door is installed at the bottom of the pipe, which is opened and closed by an SG90 servo motor.

[0009] like Figure 6 The second-layer brush cleaning mechanism includes two food-grade brush rollers, a rigid fixed shaft, and a 12V DC geared motor (600r / min, 3.6W). The two brush rollers are symmetrically arranged on both sides of the pipe, with an effective length of 180mm and an outer diameter of 90mm including the bristles. The bristles are made of food-grade nylon with a hardness of 60HA. The brush rollers are fixed to the rigid fixed shaft with 4mm set screws. The fixed shaft is 250mm long and 12mm in diameter, and its two ends are connected to the first and third layer plates via 6202 ball bearings. The input end of the fixed shaft is connected to the 12V DC geared motor via an 8-12mm coupling. The motor has a rated speed of 600r / min and a rated power of 3.6W, and is vertically fixed to the outer wall of the second-layer pipe.

[0010] like Figure 7The third-layer rotten and yellow leaf removal mechanism includes a flexible back-mounted clamp (MG996R driven), a front gear rack leaf removal assembly (SG90 driven), and a front flexible leaf removal clamp (MG996R driven, made of silicone). This mechanism utilizes the third-layer vertical pipe and the bottom sloping pipe as its carrier, and includes a flexible back-mounted clamp, a front gear rack leaf removal assembly, and a flexible leaf removal clamp. An 8×5cm installation opening is provided on the back of the pipe, housing the MG996R servo-driven back-mounted clamp with a clamping force of 17.7N. An 8×5cm working opening is provided on the front of the pipe, with an SG90 servo-driven gear rack leaf removal assembly on the outer side. The rack has a total length of 22.5cm and an effective stroke of 9.7cm. The MG996R servo-driven front flexible leaf removal clamp is fixed to the end of the rack. The contact end of the clamp is made of food-grade silicone to increase friction and prevent damage to the leaves.

[0011] like Figure 8 The fourth-layer blower packaging mechanism includes a vertical centrifugal blower (5V, 10W, 10cm outlet) and a plastic bag hanging rod feeding mechanism (hanging rod, clamping spacing 12cm, pre-stores 50 bags), employing a blower-blown bag + gravity-push bag filling method without sealing. The fourth layer uses the sloping pipe outlet of the third layer as the material inlet, and is equipped with a vertically placed centrifugal blower and plastic bag hanging rod feeding mechanism. It includes a cross-turning plate and a blower; the cross-turning plate's function is to achieve a firm connection of aluminum profiles in different directions, enhancing the stability and load-bearing capacity of structural nodes, while simplifying the complexity of multi-angle assembly. The blower's function is to blow open the packaging bags, allowing the processed bok choy to fall directly into the packaging bags to complete the packaging process.

[0012] The blower outlet has a 10cm diameter and faces the bag opening on the suspension rod mechanism. It has a rated voltage of 5V DC, a rated power of 10W, and an airflow of 0.5m³ / min. The outlet is equipped with a dustproof net with a 1mm mesh diameter. The suspension rod feeding mechanism passes through the main board of the packaging machine and is fixed at the back with wire and connected to weights. By changing different weights, the bag-removal threshold of the suspension rod mechanism is adjusted, thereby controlling the filling weight of a single bag. It is suitable for plastic packaging bags up to 15cm wide. The suspension rod mechanism pre-stores 50 packaging bags, arranged in a stacked manner. The bagging logic is as follows: the blower continuously blows open the packaging bag to be filled → the bok choy falls into the bag due to gravity → the bok choy's gravity pushes the packaging bag down from the suspension rod → the next packaging bag automatically takes its place.

[0013] II. Control System Control system: such as Figure 9Using a Raspberry Pi 5 as the core control unit (8GB RAM, 64GB storage), with a built-in boot program, and in conjunction with a PCA9685 servo control board and a 5V breadboard power module, it can identify rotten and yellow leaves of bok choy and control servo motors in linkage, with a signal response time of ≤0.1ms.

[0014] The Raspberry Pi 5 is equipped with 8GB of RAM and 64GB of storage. It has a built-in startup program and is developed based on the Linux system. After powering on, it automatically loads the YOLOv8 model and servo control program, and independently completes image acquisition, yellow leaf recognition, and servo action command sending. It does not participate in the control of DC motors and blowers and does not require the assistance of a host computer.

[0015] The PCA9685 servo control board connects to the Raspberry Pi 5 via an I2C bus, with a signal response time of ≤0.1s. Channels 0-2 control the SG90 servo (flip-door, root-cutting rack, leaf-picking rack), and channels 3-5 control the MG996R servo (first-layer fixed gripper, third-layer back fixed gripper, third-layer front leaf-picking gripper). The servo control accuracy is 1°.

[0016] The 5V breadboard power module is independently powered by the PCA9685, with an output current of 700mA and overcurrent protection. The independent drive power supply includes two switching power supplies: one powers the second-layer 12V DC geared motor (output 12V DC / 0.30A), and the other powers the fourth-layer blower (output 5V DC / 50Hz).

[0017] III. Visual Perception and Information Processing System The visual perception and information processing system includes the built-in Picamera camera module (5 megapixels, 3.21mm focal length, 30FPS) on the Raspberry Pi 5 and a lightweight YOLOv8 recognition model. The model uses rotten leaves, yellow leaves, and wormholes as independent labels, and was trained on 206 samples, achieving a recognition accuracy of ≥91.3% and a single-frame recognition time of ≤0.1ms. The pnnx toolchain from the Ultralytics library was used to convert the trained YOLOv8 model from a .pt file to a lightweight ncnn file, thereby reducing the CPU load on the Raspberry Pi 5 and achieving faster response times and servo control. This allows the Raspberry Pi 5 to achieve a window recognition frame rate of 14-16fps, significantly improving the accuracy of recognition result verification and servo control, helping to adjust the work rhythm, and enabling the simultaneous processing of multiple cabbages, greatly improving the machine's efficiency.

[0018] The Picamera camera module has 5 megapixels and a focal length of 3.21mm. It is fixed on a special shelf that is raised 10cm on the third layer, facing the detection area of ​​the bok choy inside the pipe, and shoots at a frame rate of 30FPS.

[0019] The YOLOv8 model uses rotten leaves, yellow leaves, and wormholes as three independent identification labels. It was trained on 206 samples of healthy leaves, rotten leaves, yellow leaves, and wormholes of bok choy (covering different light levels, angles, and disease severity). Data augmentation (random horizontal flipping, vertical flipping, brightness and contrast adjustment, and Gaussian blur) was used. The training, validation, and test sets were divided in an 8:1:1 ratio. The AdamW optimizer and Focal Loss loss function (based on cross-entropy) were used, and the model was trained for 500 epochs using early stopping. The model used is YOLOv8, and parameters such as cache, patience, batch, perspective, filpud, filplr, erasing, and crop_fraction are added during training. Cache caches the training set to improve subsequent training speed; patience indicates early stopping after a certain number of training epochs if the training effect doesn't change, preventing overfitting; batch specifies the number of images trained in a batch; perspective indicates the probability of adding perspective distortion during image augmentation; flipud indicates the probability of flipping the image vertically during augmentation; fliplr indicates the probability of flipping the image horizontally during augmentation; erasing indicates the probability of performing erasure operations during augmentation; and crop_fraction indicates the probability of cropping during data augmentation. These parameter settings significantly improve the accuracy of model training, enabling the model to achieve satisfactory results in the prediction task and shortening the time consumed in data acquisition. The recognition accuracy is ≥91.3%, and the single-frame recognition time is ≤0.1ms. Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. It adopts a vertical four-layer integrated structure and gravity pipeline transportation to replace the conveyor belt, which greatly reduces manufacturing costs and operating energy consumption. The equipment has a compact structure and occupies a small area. 2. Food-grade brush rollers are used for cleaning to avoid residual moisture accelerating spoilage. The cost of a single brush roller mechanism is only about 285 yuan, which is about 1 / 40 of the cost of traditional bubble pool cleaning solutions, greatly reducing the equipment manufacturing cost. 3. The cleaning process is placed before visual recognition to eliminate interference from dirt and dust, and the YOLOv8 model is used to achieve accurate identification of rotten and yellow leaves (accuracy ≥ 91.3%). 4. The device uses dual flexible grippers in conjunction with a gear and rack mechanism to simulate the tearing and removal of rotten and yellow leaves by hand, thus avoiding damage to healthy leaves; 5. It adopts a bag-blowing method with a blower and gravity pushing, and the suspension rod mechanism automatically changes bags. The structure is simple and the packaging efficiency is high. 6. The control system has a built-in automatic startup program, requiring no host computer assistance, and operates fully automatically, with a single plant processing time of ≤10s. 7. This invention adopts a modular design approach to process bok choy. Each process can be implemented in parallel using relatively simple methods, and even the machine itself can be connected in parallel, which greatly improves the efficiency of processing bok choy. Detailed Implementation

[0020] The specific implementation method of the vertical bok choy purification and packaging integrated machine is as follows.

[0021] When the device is powered on, the Raspberry Pi 5 automatically loads the YOLOv8 model and servo control program, and the 12V DC geared motor and blower run continuously, while the device is in standby mode.

[0022] The Raspberry Pi 5 triggers the MG996R retaining jaws of the first layer to close via the PCA9685, thus axially fixing the bok choy. The SG90 servo drives the root-cutting rack to reciprocate, causing the root-cutting blade to complete the root cutting at the 3.5cm working position. The rack then resets, and the retaining jaws release.

[0023] The first-level SG90 servo flip door opened, and the little cabbage fell to the second-level pipe due to gravity.

[0024] A 12V DC geared motor drives two brush rollers to rotate relative to each other, clamping and cleaning the bok choy, brushing away surface impurities and weak, rotten, yellow leaves. After cleaning, the bok choy falls to the third-level sloping pipe due to gravity.

[0025] The Raspberry Pi 5 triggers the closure of the third-layer MG996R back-mounted gripper, the Picamera captures an image of the bok choy, and the Raspberry Pi 5 runs the YOLOv8 model to identify rotten / yellow leaves. If a target is identified, a command is sent to the PCA9685. The specific control process for identifying rotten / yellow leaves is as follows: Figure 10 As shown.

[0026] The SG90 servo drives the leaf-removing rack forward, and the MG996R front leaf-removing gripper reaches the position of the rotten yellow leaf, simulating a human hand tearing it off, and the rack returns to its original position. The third-layer back-fixing gripper releases, and the bok choy falls down the sloping pipe to the fourth layer by gravity.

[0027] The fourth-layer blower continuously blows open the plastic packaging bags on the suspension rod mechanism, and the bok choy falls into the bags. When the weight of the bok choy in the bag reaches 500g, gravity pushes the packaging bag away from the suspension rod, completing the unsealed bagging. The bagged bok choy falls into the finished product receiving box, and the suspension rod mechanism automatically replaces it with the next packaging bag. The blower continues to blow open the bags, entering the next processing cycle.

[0028] The entire processing time for a single bok choy plant is ≤10 seconds, requiring no manual intervention throughout the process. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall structure of the present invention. Figure 2 This is a schematic diagram of the fuselage frame system structure of the present invention. Figure 3 This is a schematic diagram of the corner groove of the present invention. Figure 4 This is a schematic diagram of the gravity transmission pipeline structure of the present invention. Figure 5 This is a schematic diagram of the gear rack and tool fixture of the present invention. Figure 6 This is a schematic diagram of the brush roller structure of the present invention. Figure 7 This is a schematic diagram of the flexible gripper structure of the present invention. Figure 8 This is a schematic diagram of the packaging mechanism of the present invention. Figure 9 Right view of the packaging mechanism of the present invention Figure 10 This is a schematic diagram of the visual recognition and yellow leaf removal system of the present invention. Figure 11 This is a control flowchart of the control system and visual perception system of the present invention.

Claims

1. A vertical integrated machine for purifying, processing, and packaging bok choy based on computer vision, characterized in that, This includes motion execution systems, control systems, and visual perception and information processing systems; The motion execution system integrates a frame, a flexible gripper mechanism, a gear and rack mechanism, a flap door, a brush, and a packaging mechanism to perform mechanical actions such as cutting the roots, cleaning, removing rotten and yellow leaves, and packaging the bok choy into unsealed bags. The frame is constructed from 20×20×R1.0 industrial aluminum profiles and secured by angle brackets, bolts, nuts, and aluminum profile corner groove connectors. The flexible gripper mechanism is characterized by a gripper frame, a servo gear housing, a toothed connecting rod, a connecting rod, a gripper fixing component, and a flexible gripper. The gripper frame is fixed at one end to a pipe wall or rack platform and at the other end to the flexible gripper. The flexible gripper is a 3D printed component with internal reinforcing ribs of 1mm thickness every 5mm along a 45° direction, allowing the gripper to adapt to the shape of the object being gripped. The gear and rack mechanism includes a servo gear bracket, gears, and a rack. An SG90 servo gear is fixed to the back of the servo gear bracket, and the front uses a large-diameter 4mm... M-bolts secure the gear to the servo motor, and the same bolts secure the servo motor bracket to the shelf. The SG90 servo motor outputs a 0-180° rotation angle, which is converted into linear motion via a gear and rack mechanism, achieving an effective stroke of 97mm. The flap door includes a servo motor frame and a 50mm x 50mm food-grade PP board. Bolts secure the SG90 servo motor to the flap door PP board. A 95° rotation of the servo motor opens the flap door. The brush mechanism includes a 180mm long, 12mm inner diameter food-grade brush roller, a 250mm long, 12mm diameter 45 steel chrome-plated hard shaft, an 8-to-12 coupling, a geared DC motor bracket, and a 12mm diameter bearing. The brush roller and hard shaft are secured by 4mm set screws. The hard shaft is secured to the first and second shelf boards by bearings. The geared DC motor bracket is secured to the outer wall of the first layer pipe by 4mm bolts. The packaging mechanism includes a blower and a feeding mechanism for unsealed bagging of the bok choy. The control system is based on a Raspberry Pi 5 and is responsible for the overall automated operation logic and servo motor linkage control of the device. It includes a Raspberry Pi 5, a PCA9685 servo control board, and a 5V breadboard power module. The Raspberry Pi is used to carry a Picamera camera module and output PWM signals to the PCA9685 servo control board to control 16 servos. The 5V breadboard power module is connected to a 9V 1.5A power adapter to power the PCA9685 control board through the breadboard. The visual perception and information processing system includes the Picamera camera module built into the Raspberry Pi 5 and a loaded YOLOv8 lightweight model, used to acquire images of bok choy and identify rotten and yellow leaves in real time; the Picamera camera module is a Picamera... The V1.3 camera module is highly compatible with Raspberry Pi, featuring a 5-megapixel sensor. Equipped with a lightweight Yolov8 model, it achieves 14-16fps. This invention uses a dataset of 160 images (train, validation, and test) in an 8:1:1 ratio, with samples of rotten and yellowed cabbage leaves and wormholes. For model training, random perspective transformation, vertical and horizontal inversion are added to increase data diversity. Under certain probability, random image erasure simulates occlusion to aid model robustness. When cropping images, all content is retained to maintain image integrity. From image acquisition to outputting recognition results and exporting data, a single complete recognition process takes approximately 3.4 seconds, including 4.5ms for preprocessing the predicted image, 0.1ms for inference, and 1.0ms for image processing. To ensure prediction accuracy, the program can output prediction results and bounding box information for debugging. In practical applications, only data export is needed, further reducing the time required. The equipment has a vertical four-layer integrated structure, which is suitable for the integrated pretreatment of small Chinese cabbage with a stem diameter of 5cm.

2. The vertical bok choy purification and packaging integrated machine based on computer vision according to claim 1, characterized in that, The motion execution system includes a fuselage frame, a gravity transmission pipeline, a first-layer root-cutting mechanism, a second-layer brush cleaning mechanism, a third-layer rotten and yellow leaf removal mechanism, and a fourth-layer blower packaging mechanism. The frame is constructed from 20×20×R1.0 industrial aluminum profiles and is fixed by corner brackets, bolts, nuts and aluminum profile corner groove connectors. The frame has four vertical columns with a column spacing of 355×355mm. The heights of the four working areas are 150mm, 573mm and 369mm respectively. The gravity transmission pipeline is a rigid, food-grade PP vertical pipeline that runs through four layers. The inner cross-section of the pipeline is 5cm×7cm and the wall thickness is 5mm. Each layer of pipeline is arranged vertically and coaxially, with a flared transition structure on all four sides. The transition opening diameter is 10cm. The bottom of the third layer of pipeline is a sloping pipeline that is seamlessly connected to the fourth layer.

3. The vertical bok choy purification and packaging integrated machine based on computer vision according to claim 2, characterized in that, The first-layer root cutting mechanism utilizes a material hopper, flexible fixed grippers, a gear and rack root cutting assembly, and an SG90 servo motor to control the flap door. The feeding hopper is inverted cone-shaped with an upper opening of 10×10cm, and is seamlessly connected to the top of the first layer of pipe; The flexible fixing claw is located at the mounting port on the back of the pipe (8×5cm), driven by an MG996R servo motor, with a clamping stroke of 1.5-72mm and a clamping force of 17.7N; The gear and rack root cutting assembly is located at the root cutting working port (6×4cm) on the front of the pipe, with the working port 3.5cm from the bottom of the pipe. The SG90 servo motor drives the rack to reciprocate. The rack has a total length of 15cm and an effective stroke of 9.7cm. The end is fixed with a 3mm thick hard stainless steel root cutting blade by a 4mm bolt. The flap door is rectangular, with an area of ​​55mm x 45mm, and its opening and closing are controlled by an SG90 servo motor.

4. The vertical bok choy purification and packaging integrated machine based on computer vision according to claim 2, characterized in that, The second layer of the brush cleaning mechanism includes two food-grade brush rollers, a rigid fixed shaft, a coupling, and a 12V DC geared motor; The brush rollers are symmetrically arranged on both sides of the pipe, with an effective length of 180mm and an outer diameter of 90mm including the bristles. The bristles are made of food-grade nylon material with a hardness of 60HA. The rigid fixed shaft is 250mm long and 12mm in diameter. The brush roller is fixed to the shaft by a 4mm set screw. Both ends of the fixed shaft are connected to the shelf through 6202 bearings. The input end is connected to a 12V DC geared motor through an 8-12mm plum blossom coupling. The motor has a rated speed of 600r / min and a rated power of 3.6W.

5. The vertical bok choy purification and packaging integrated machine based on computer vision according to claim 2, characterized in that, The third layer of rotten and yellowed leaf removal mechanism includes a flexible fixing gripper on the back, a gear and rack leaf removal assembly on the front, and a flexible leaf removal gripper. The flexible fixing claw on the back is located at the mounting port on the back of the pipe (8×5cm), driven by an MG996R servo motor, with a clamping force of 17.7N; The front gear rack leaf-removing assembly is located at the front working port of the pipeline (8×5cm). The SG90 servo motor drives the rack to reciprocate linearly. The rack has a total length of 22.5cm and an effective stroke of 9.7cm. The front flexible leaf-removing gripper driven by the MG996R servo motor is fixed at the end of the rack. The contact end of the gripper is made of food-grade silicone.

6. The vertical bok choy purification and packaging integrated machine based on computer vision according to claim 2, characterized in that, The fourth-layer blower packaging mechanism includes a vertically placed centrifugal blower and a feeding mechanism; The blower has an air outlet diameter of 10cm, which is directly opposite the opening of the packaging bag on the suspension rod mechanism. It has a rated voltage of 5V DC, a rated power of 10W, and an air volume of 0.5m³ / min. The air outlet is equipped with a dustproof net with a mesh diameter of 1mm. The feeding mechanism is equipped with a suspension rod for placing packaging bags. The rod passes through the main board of the packing machine, is fixed at the back with wire, and is connected to weights. By changing weights of different weights, the bag removal threshold of the suspension rod mechanism is adjusted, thereby controlling the filling weight of a single bag. It is suitable for plastic packaging bags with a width of 15cm. The feeding mechanism has 50 packaging bags pre-stored and is arranged in a stacked manner. The bagging method uses a blower to blow the bag and the bok choy to push the bag with gravity. After the bag is filled, the gravity of the bok choy pushes the bag away from the hanging rod, and the next bag automatically takes its place.

7. The vertical bok choy purification and packaging integrated machine based on computer vision according to claim 1, characterized in that, The control system uses Raspberry Pi 5 as the core main control unit, in conjunction with PCA9685 servo control board and 5V breadboard power module. The Raspberry Pi 5 is equipped with 8GB of RAM and 64GB of storage. It has a built-in startup program and is developed based on the Linux system. After powering on, it automatically loads the YOLOv8 model and servo control program, and independently completes image acquisition, yellow leaf recognition and servo action command sending without the need for an external host computer. The PCA9685 servo control board is connected to the Raspberry Pi 5 via an I2C bus, with a signal response time of ≤0.1s. Channels 0-2 control the SG90 servo, and channels 3-5 control the MG996R servo. The 5V breadboard power module outputs a current of 700mA and is equipped with overcurrent protection.

8. The vertical bok choy purification and packaging integrated machine based on computer vision according to claim 1, characterized in that, The visual perception and information processing system includes the Picamera camera module that comes with the Raspberry Pi 5 and the loaded YOLOv8 lightweight recognition model. The Picamera camera module has 5 megapixels, a focal length of 3.21mm, and is positioned directly in front of the detection area for the bok choy inside the third-layer pipe, with a shooting frame rate of 30FPS. The YOLOv8 model uses rotten leaves, yellow leaves, and wormholes as three independent recognition labels. It was trained on 206 samples of healthy leaves, rotten leaves, yellow leaves, and wormholes of bok choy, achieving a recognition accuracy of ≥91.3% and a single-frame recognition time of ≤0.1ms.

9. The vertical bok choy purification and packaging integrated machine based on computer vision according to any one of claims 1 to 8, characterized in that, The overall automated operation logic of the equipment is as follows: The bok choy enters the first-layer pipe from the top hopper → Raspberry Pi 5 triggers the MG996R fixing gripper on the first layer to close via PCA9685 → SG90 servo drives the cutting rack to complete the cutting → The rack resets and the fixing gripper releases → The SG90 servo's flip door on the first layer opens, and the bok choy falls along the pipe to the second layer by gravity → A 12V DC geared motor drives the brush roller to rotate, cleaning the surface of the bok choy → After cleaning, the bok choy falls by gravity to the third-layer sloping pipe → Raspberry Pi 5 triggers the MG996R back fixing gripper on the third layer to close → Picamera captures images, YOLOv8 model identifies rotten / yellow leaves → After target identification, SG90 servo drives leaf-removing rack forward, MG996R front leaf-removing gripper simulates human hand tearing to remove leaves → rack resets, back gripper releases → bok choy falls down the inclined pipe to the fourth layer by gravity → blower continuously blows open the packaging bag on the suspension rod mechanism, bok choy falls into the bag → bok choy gravity pushes the packaging bag away from the suspension rod mechanism, the next packaging bag automatically takes its place → bagged bok choy falls into the finished product receiving box; the entire process time for a single bok choy is ≤10s.

Images