A device and method for detecting inner cardboard in irregularly shaped packaging machines

Abstract

This invention discloses an inner cardboard detection device and method for irregularly shaped packaging machines, belonging to the field of detection technology. The device includes a transmission mechanism, an imaging chamber, a control chamber, a rejection chamber, and a human-machine interface. The control chamber is connected to the human-machine interface, transmission mechanism, imaging chamber, and rejection chamber via wiring. The method uses six cameras to capture images of six sides of the inner cardboard package and utilizes a fusion of traditional vision and deep learning to achieve high-efficiency, high-quality inner cardboard packaging quality detection, enabling online rejection. It effectively balances detection accuracy and debugging efficiency, providing strong support for production and possessing broad application potential.

Description

Technical Field

[0001] This invention belongs to the field of detection technology, specifically relating to an inner cardboard detection device and method for irregularly shaped packaging machines. Background Technology

[0002] With the continuous development of the industry, the application of irregularly shaped cigarettes and cigarette packs is increasing, leading to a greater diversity in the specifications of the cardboard used in cigarette packaging. A typical cigarette pack usually consists of cigarettes, a cigarette box, inner cardboard, and aluminum foil (or insulating paper). The inner cardboard plays a crucial role, fixing the shape of the cigarettes and providing support, thus becoming an indispensable part of the cigarette pack packaging. During the packaging process, the cigarettes are arranged according to specific specifications, and then the inner cardboard is bonded together using hot melt adhesive. However, problems such as missing inner cardboard, hot melt adhesive overflow, loose packaging, improper bonding, and surface damage frequently occur during the inner cardboard packaging process. These defective inner cardboard packs are difficult to detect in subsequent processes, leading to serious product quality issues.

[0003] Currently, visual inspection equipment is widely used, and it can be divided into traditional visual inspection and deep learning inspection. Traditional visual inspection equipment uses rule-based detection, setting the detection area, algorithm type, and detection parameters. Its advantages are mature and transparent algorithms and high efficiency. Its disadvantages are that it relies heavily on the experience of the debugging personnel and cannot achieve full coverage detection of internal paper jams, potentially missing detections in uncovered areas. Deep learning inspection methods offer higher accuracy and better stability, but require a large number of sample images, especially defect images. Therefore, the installation and debugging cycle is long, generally several weeks to a month, and a certain number of defect images need to be collected to ensure model accuracy.

[0004] Therefore, developing a device capable of accurately and efficiently detecting defects in inner cardboard during the cigarette packaging process and achieving online rejection of defective inner cardboard has become an urgent technical challenge to be solved in this field. Solving this challenge will help ensure the quality and consistency of cigarette packaging. Summary of the Invention

[0005] In view of the above-mentioned technical problems in the prior art, the present invention proposes an inner cardboard detection device and method for irregular packaging machines. The detection method organically integrates traditional vision and deep learning detection methods. It is reasonably designed, overcomes the shortcomings of the prior art, and has a better and more efficient detection effect.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] An inner cardboard detection device for a non-standard packaging machine includes a conveying mechanism, a shooting chamber, a control chamber, a rejection chamber, and a human-machine interface; the control chamber is connected to the human-machine interface, the conveying mechanism, the shooting chamber, and the rejection chamber via wiring.

[0008] The transmission mechanism is configured to enable the internal card packet to move and be transmitted in a defined direction;

[0009] The shooting chamber is configured to capture images of all six sides of the inner card holder;

[0010] The control compartment is configured for image processing, motor control, rejection valve control, and communication with the packaging machine cabinet.

[0011] The rejection bin is configured for online rejection and collection of defective card packets;

[0012] The human-computer interface is configured for human-computer interaction.

[0013] Preferably, the transmission mechanism includes a servo motor, a coupling, a synchronous pulley, a driven pulley, an idler pulley, a back tension pulley, a belt, an inlet guide plate, an inlet support plate, an outlet guide plate, an outlet support plate, and a wire rope;

[0014] The servo motor drives the synchronous pulley via a coupling to achieve power transmission; the upper side of the synchronous pulley is the tight side, and the lower side is the loose side; the inlet support plate and outlet support plate are located below the tight side of the synchronous pulley and are used to support the synchronous pulley;

[0015] The first back tensioning pulley, the first idler pulley, the second back tensioning pulley, the second idler pulley, and the third back tensioning pulley are located on the slack side of the timing belt pulley. They not only support the timing belt pulley but also control the tension of the timing belt pulley by adjusting the position of the second idler pulley.

[0016] The function of the inlet guide plate, outlet guide plate, and wire rope is to restrict the movement direction of the inner bag during cigarette pack transportation, ensuring that it moves along the required path;

[0017] The wire rope is made of stainless steel wire with a diameter of 1 mm.

[0018] Preferably, the control compartment includes a vision-specific controller, a core control board, a power control module, and an emergency stop control module;

[0019] A dedicated vision controller, equipped with a CPU and GPU, is configured to handle vision-related computations;

[0020] The core control board includes a core control unit, a camera communication interface, a light source communication interface, input / output interfaces, and related circuits;

[0021] The core control board communicates with the vision-dedicated controller via a serial port and is configured to be responsible for controlling the camera and light source. When the image sensor detects the internal card packet, the core control board lights up the light source and triggers the camera to take a picture.

[0022] The core control board is responsible for communication with the electrical cabinet and control of the rejection valve. It controls the operation of the electromagnetic rejection valve in the rejection compartment based on the calculation results of the vision controller, receives the status information of the electrical cabinet, and outputs the statistical data to the electrical cabinet.

[0023] The core control board is also responsible for controlling the servo motor, controlling the speed of the servo motor to exceed the running speed of the inner card pack conveyor belt, realizing the effective separation of the inner card pack, ensuring that each inner card pack is well isolated when it arrives at the shooting chamber, controlling the start and stop adjustment of the servo motor, and realizing linkage and synchronous start and stop with the entire packaging machine;

[0024] The power control module includes a load disconnect switch, a filter, and a switching power supply;

[0025] The load disconnect switch is the main power switch for the entire system and is configured to control the on / off of the external 220V power supply.

[0026] The filter is configured to remove noise and spurious signals from the alternating current to ensure a stable and clean voltage for the system power supply.

[0027] The switching power supply is configured to convert 220V power to 24V power to supply the power required by the vision-specific controller and the core control board.

[0028] The emergency stop control module is configured to control the start and stop of the motor in emergency or abnormal situations.

[0029] Preferably, the shooting chamber includes 6 sets of Ethernet industrial cameras, a ring LED light source, and a photo sensor;

[0030] The entire shooting chamber adopts a spherical integrating optical path structure, and its inner surface is covered with white matte paint;

[0031] The six Ethernet industrial cameras are a top camera, a bottom camera, a front camera, a rear camera, a left-side camera, and a right-side camera; each Ethernet industrial camera is equipped with a low-distortion lens.

[0032] The ring-shaped LED light source includes an upper light source and a lower light source. The ring-shaped LED light source is fixed in the middle of the sphere. The upper and lower light sources are arranged symmetrically. The upper and lower light sources do not directly illuminate the inner cardboard, but are diffusely reflected to the surface of the inner card pack through the inner surface of the shell, ensuring that all six sides of the entire inner card pack are uniformly illuminated.

[0033] The shooting compartment has a circular hole the size of a camera lens at the center of the top, bottom, front, and back for taking pictures; a photo sensor port is located on the side of the camera hole at the top of the shooting compartment for the photo sensor to determine the position of the internal card holder; there is a rectangular hole at the center of the left and right sides for the internal card holder, belt, and limiting wire to pass through, and also for taking pictures.

[0034] When the inner card bag reaches the designated position, the image sensor is triggered, and six sets of Ethernet industrial cameras capture images of the six sides of the inner card bag and transmit them to the controller for processing.

[0035] Preferably, the rejection chamber includes an electromagnetic rejection air valve, a booster air nozzle, two rejection sensors, a rejection cover, and a rejection box;

[0036] The booster nozzle and the rejection hood are located on both sides of the belt, with the booster nozzle aligned with the inlet of the rejection hood.

[0037] The rejection box has a rectangular frame structure and is located below the rejection hood. It is used to collect defective inner card packs. The bottom of the rejection box is equipped with four rollers for pulling out.

[0038] The two rejection sensors are symmetrically placed reflective sensors, located on both sides of the booster nozzle. When the inner bag reaches the rejection position and both rejection sensors are blocked at the same time, if the controller's detection result is unqualified, the electromagnetic rejection valve opens, and the booster nozzle blows the defective inner bag into the rejection hood, where it will then slide into the rejection box.

[0039] Preferably, the human-machine interface includes a touch LCD screen and a rocker arm;

[0040] The rocker arm adopts a three-axis structure, which enables it to be adjusted in three dimensions. The rocker arm is fixed to the side of the detection device with screws. Users can browse relevant information about the system operation through the touch screen and set various parameters of the system.

[0041] Furthermore, this invention also mentions a method for detecting inner paper jams on a non-standard packaging machine. This method employs an inner paper jam detection device for a non-standard packaging machine as described above, and includes the following implementation steps:

[0042] Step 1: Quick Start Phase;

[0043] Initially, traditional vision methods were used to quickly detect and identify most common defects;

[0044] Step 2: Collection and classification of sample data;

[0045] The sample images of each camera are classified as qualified and unqualified. Unqualified samples are further classified into several categories of defects, including damage, missing parts, glue overflow, loose packaging, and inadequate bonding.

[0046] Step 3: Training and enabling the deep learning model;

[0047] For the collected positive and negative sample data from the 6 cameras, a deep learning classification model is trained for each camera. After obtaining the exclusive classification model for each camera, the deep learning detection function is enabled.

[0048] Step 4: Iterative optimization of the model;

[0049] Add invalid sample data and iterate the deep learning classification model for each camera.

[0050] Preferably, step 3 specifically includes the following steps:

[0051] Step 3.1: Model Building;

[0052] Step 3.2: Dedicated model training;

[0053] Step 3.3: Model deployment enabled.

[0054] Preferably, step 4 specifically includes the following steps:

[0055] Step 4.1: Model format conversion;

[0056] Step 4.2: Import the dynamic link library;

[0057] Step 4.3: Read in the category file;

[0058] Step 4.4: Load the model file;

[0059] Step 4.5: Convert the image into an input tensor;

[0060] Step 4.6: Set the model input;

[0061] Step 4.7: Perform network inference;

[0062] Step 4.8: Obtain the reasoning results;

[0063] Step 4.9: Iterative optimization of the model.

[0064] The beneficial technical effects of this invention are as follows:

[0065] The entire device uses six cameras to capture images of six sides of the inner card pack and utilizes a fusion of traditional vision and deep learning to achieve high-efficiency and high-quality inner card pack quality inspection. It can also reject items online, achieving a good balance between inspection accuracy and debugging efficiency, providing strong support for production, and has a wide range of applications. Attached Figure Description

[0066] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0067] Among them, 1-transmission mechanism; 2-shooting chamber; 3-rejection chamber; 4-control chamber; 5-human-machine interface; 6-alarm device;

[0068] Figure 2 This is an electrical schematic diagram of the present invention;

[0069] Figure 3 This is a schematic diagram of the transmission mechanism structure;

[0070] Among them, 1-1-Inlet guide plate; 1-2-Inlet support plate; 1-3-Inner clamp; 1-4-Wire rope; 1-5-First sensor; 1-6-Air nozzle; 1-7-Second sensor; 1-8-Outlet support plate; 1-9-Outlet guide plate; 1-10-Synchronous belt pulley; 1-11-Coupling; 1-12-Servo motor; 1-13-First back tension pulley; 1-14-First idler pulley; 1-15-Second back tension pulley; 1-16-Second idler pulley; 1-17-Synchronous belt; 1-18-Third back tension pulley; 1-19-Driven pulley;

[0071] Figure 4 This is a schematic diagram of the warehouse structure.

[0072] Among them, 2-1-top camera; 2-2-image sensor port; 2-3-front camera; 2-4-left side camera; 2-5-lower light source; 2-6-bottom camera; 2-7-inner card slot; 2-8-rear camera; 2-9-upper light source; 2-10-right side camera;

[0073] Figure 5 This is a schematic diagram of the removal warehouse structure;

[0074] Among them, 3-1-rejection hood; 3-2-first rejection sensor; 3-3-boosting nozzle; 3-4-second rejection sensor; 3-5-electromagnetic rejection valve; 3-6-rejection box;

[0075] Figure 6 This is a schematic diagram of the control warehouse structure;

[0076] Among them, 4-1-load disconnect switch; 4-2-emergency stop control module; 4-3 filter; 4-4 switching power supply; 4-5-core control board; 4-6-vision dedicated controller; 4-7-motor driver;

[0077] Figure 7 A schematic diagram of the human-computer interface structure;

[0078] Among them, 5-1-Front side of HMI; 5-2-USB interface; 5-3-Power switch; 5-4-Rocker arm; 5-5-Back side of HMI;

[0079] Figure 8 This is a flowchart of the detection and processing procedure. Detailed Implementation

[0080] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0081] A device for detecting inner paper jams on irregularly shaped packaging machines, the overall structure of which is as follows: Figure 1 As shown, the principle is as follows Figure 2 As shown, the detection device includes a transmission mechanism 1, an imaging chamber 2, a control chamber 4, a rejection chamber 3, and a human-machine interface 5.

[0082] like Figure 3 As shown, the transmission mechanism 1 consists of a servo motor 1-12, a coupling 1-11, a synchronous pulley 1-10, a driven pulley 1-19, a first idler pulley 1-14, a second idler pulley 1-16, a first back tension pulley 1-13, a second back tension pulley 1-15, a third back tension pulley 1-18, a belt, an inlet guide plate 1-1, an inlet support plate 1-2, an outlet guide plate 1-9, an outlet support plate 1-8, and a wire rope 1-4. The servo motor 1-12 drives the synchronous pulley 1-10 via the coupling 1-11 to achieve power transmission. The upper side of the synchronous pulley 1-10 is the tight side, and the lower side is the loose side. The inlet support plate 1-2 and the outlet support plate 1-8 are located below the tight side of the synchronous pulley 1-10, and their function is to support the synchronous pulley 1-10. The first rear tensioning pulley 1-13, the first idler pulley 1-14, the second rear tensioning pulley 1-15, the second idler pulley 1-16, and the third rear tensioning pulley 1-18 are located on the slack side of the synchronous belt pulley 1-10. They not only support the synchronous belt pulley 1-10 but also control the tension of the synchronous belt pulley 1-10 by adjusting the position of the idler pulley 1-16. The inlet guide plate 1-1, the outlet guide plate 1-9, and the wire rope 1-4 restrict the movement direction of the inner card pack 1-3 during cigarette pack transportation, ensuring that it moves along the required path. The wire rope 1-4 is made of 1 mm diameter stainless steel wire, which has minimal impact on obstructing the inner card pack 1-3 and will not significantly affect camera shooting.

[0083] like Figure 4As shown, the imaging chamber 2 is composed of the following components: six Ethernet industrial cameras, six low-distortion lenses, two ring-shaped high-brightness LED light sources, and an image sensor. The six Ethernet industrial cameras are designated as top camera 2-1, bottom camera 2-6, front camera 2-3, rear camera 2-8, left side camera 2-4, and right side camera 2-10. Each Ethernet industrial camera is equipped with a low-distortion lens. The ring-shaped LED light sources include an upper light source 2-9 and a lower light source 2-5. The ring-shaped LED light sources are fixed in the center of the sphere. The upper light source 2-9 and the lower light source 2-5 are symmetrically arranged. The upper light source 2-9 and the lower light source 2-5 do not directly illuminate the inner cardboard, but rather diffusely reflect light onto the surface of the inner cardboard package through the inner surface of the housing, ensuring that all six sides of the entire inner cardboard package receive uniform illumination and avoiding localized overexposure. At the center of the top, bottom, front, and back of the shooting chamber 2, there is a circular hole the size of a camera lens diameter for camera shooting. On the side of the camera hole at the top of the shooting chamber 2, there is a photo sensor port 2-2 for the photo sensor to determine the position of the inner card bag. There is a rectangular hole at the center of the left and right sides for the inner card bag 1-3, the belt, and the limiting steel wire rope 1-4 to pass through, and also for camera shooting. When the inner card bag 1-3 reaches the designated position, the photo sensor is triggered, and 6 sets of Ethernet industrial cameras capture images of the six sides of the inner card bag and transmit them to the controller for processing.

[0084] The entire shooting chamber adopts a spherical integrating optical path structure, and its inner surface is covered with white matte paint.

[0085] like Figure 6 As shown, the control compartment includes a vision-dedicated controller 4-6, a core control board 4-5, a power control module, and an emergency stop control module 4-2; it is used to implement various control and monitoring functions to ensure the normal operation and safety of the system.

[0086] Control Module 4 is divided into the following four main parts:

[0087] Dedicated Vision Controller: This controller is equipped with a high-performance CPU and GPU to meet the needs of traditional real-time vision detection as well as the online training and inference of deep learning models. Its task is to handle vision-related computations.

[0088] Core Control Board: The core control board consists of a core control unit, camera communication interface, light source communication interface, input / output interfaces, and related circuits. It communicates with the dedicated vision controller via serial port and is responsible for controlling the camera and light source. When the image sensor detects an inner card pack, it illuminates the light source and triggers the camera to take a picture. In addition, the core control board is responsible for communication with the electrical cabinet and controlling the rejection valve. Based on the calculation results of the dedicated vision controller, it controls the operation of the rejection valve, receives electrical cabinet status information, and outputs statistical data to the electrical cabinet. Furthermore, the core control board is also responsible for controlling the servo motor, ensuring its speed exceeds the running speed of the inner card pack conveyor belt, achieving effective separation of the inner card packs and ensuring that each inner card pack is well isolated upon arrival at the imaging chamber; it also controls the start and stop of the servo motor, achieving linkage and synchronous start and stop with the entire packaging machine.

[0089] Power Control Module: The power control module includes a load disconnect switch 4-1, a filter 4-3, and a switching power supply 4-4. The load disconnect switch 4-1 is the main power switch for the entire system, used to control the on / off state of the external 220V power supply. The filter 4-3 removes noise and interference from the AC power supply to ensure a stable and clean voltage for the system. The switching power supply 4-4 converts the 220V power supply to 24V to power the vision-dedicated controller 4-6 and the core control board 4-5.

[0090] Emergency stop control module: The emergency stop control module is used to control the start and stop of the motor in emergency or abnormal situations.

[0091] like Figure 5 As shown, the rejection chamber 2 includes an electromagnetic rejection air valve 3-5, a booster air nozzle 3-3, a first rejection sensor 3-2, a second rejection sensor 3-4, a rejection cover 3-1, and a rejection box 3-6.

[0092] The booster nozzle 3-3 and the rejection hood 3-1 are distributed on both sides of the belt, with the booster nozzle 3-3 aligned with the inlet of the rejection hood 3-1; the rejection box 3-6 adopts a rectangular frame structure and is located below the rejection hood 3-1, used to collect the inner card bag 1-3 of defects. The bottom of the rejection box 3-6 is equipped with four rollers for pulling operation.

[0093] The first rejection sensor 3-2 and the second rejection sensor 3-4 are symmetrically placed reflective sensors, located on both sides of the booster nozzle 3-3. When the inner bag 1-3 reaches the rejection position and the first rejection sensor 3-2 and the second rejection sensor 3-4 are blocked at the same time, if the controller's detection result is unqualified (NG), the electromagnetic rejection air valve 3-5 opens, and the booster nozzle 3-3 blows the defective inner bag 1-3 into the rejection cover 3-1, where it will then slide into the rejection box 3-6.

[0094] Compared to designs using a single sensor, employing two sensors allows for more accurate determination of the inner card's position, ensuring that the pressurized air nozzle is always precisely aligned with the center of the inner card. This design effectively addresses varying packaging machine speeds, preventing misaligned or skewed airflow and thus improving the accuracy and efficiency of rejection.

[0095] like Figure 7 As shown, the human-machine interface 5 includes a touch LCD screen, an industrial control board, a housing, and a rocker arm 5-4; it is used to run human-machine interaction software.

[0096] The rocker arm 5-4 adopts a three-axis structure, which enables it to be adjusted in three dimensions. The back of the human-machine interface 5-5 is equipped with four screws for connecting the rocker arm 5-4. The rocker arm 5-4 is fixed to the side of the detection device by the screws. Users can browse relevant information about the system operation through the touch screen and set various parameters of the system.

[0097] A method for detecting inner paper jams on irregularly shaped packaging machines employs a combined traditional vision and deep learning approach to better identify both smaller and more complex defects. The flowchart of its detection process is shown below. Figure 8 As shown, the implementation steps include the following:

[0098] S1. Rapid Start-up Phase of Traditional Vision Methods. In the initial stage, conventional vision inspection algorithms are used, which can quickly detect and identify most common defects, allowing the inspection equipment to be quickly put into use on the production line. For example, integrity detection algorithms can be used to identify defects such as damaged or missing inner paper, contour calculation and pixel counting algorithms can identify glue overflow and loose packaging, and edge detection algorithms can identify problems with incomplete bonding.

[0099] S2. Sample Data Collection and Classification. To establish an effective defect detection system, sample data needs to be collected. For six different cameras, acceptable (OK) and unacceptable (NG) samples are collected, and these NG samples are classified in detail according to the type of defect, such as breakage, missing parts, excess glue, loose parts, and inadequate bonding. This step ensures a diverse range of data for training.

[0100] S3. Training and Enabling of Deep Learning Models. Using the collected positive and negative sample data, deep learning classification models will be trained camera-by-camera, with a customized model for each camera to ensure the model fully understands the data characteristics of each camera. After model training is complete, the deep learning detection function will be enabled to identify smaller and more complex defects, improving the system's detection accuracy.

[0101] S4. Iterative Optimization of the Model. Improving a defect detection system is a gradual process. To enhance the model's accuracy and precision, the deep learning classification model for each camera is repeatedly optimized and iterated. This includes continuing to train the model to adapt to new data and further improving its performance to ensure the system maintains a high level of efficiency and accuracy over long-term operation.

[0102] By combining traditional vision with deep learning, the method not only improves the shortcomings of traditional vision detection accuracy but also overcomes the impact of long debugging cycles in deep learning. It can accurately and efficiently detect various packaging defects on inner cardboard and achieve online rejection, thus achieving a good balance between detection accuracy and debugging efficiency.

[0103] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for detecting inner cardboard in a non-standard packaging machine, characterized in that: An inner paper jam detection device for a non-standard packaging machine is adopted. The device includes a transmission mechanism, a shooting chamber, a control chamber, a rejection chamber, and a human-machine interface. The control chamber is connected to the human-machine interface, the transmission mechanism, the shooting chamber, and the rejection chamber via wiring. The transmission mechanism is configured to enable the internal card packet to move and be transmitted in a defined direction; The shooting chamber is configured to capture images of all six sides of the inner card holder; The control compartment is configured for image processing, motor control, rejection valve control, and communication with the packaging machine cabinet. The rejection bin is configured for online rejection and collection of defective card packets; The human-computer interface is configured for human-computer interaction. The transmission mechanism includes a servo motor, coupling, synchronous pulley, driven pulley, idler pulley, back tensioner, belt, inlet guide plate, inlet support plate, outlet guide plate, outlet support plate, and wire rope; The servo motor drives the synchronous pulley via a coupling to achieve power transmission; the upper side of the synchronous pulley is the tight side, and the lower side is the loose side; the inlet support plate and outlet support plate are located below the tight side of the synchronous pulley and are used to support the synchronous pulley; The first back tensioning pulley, the first idler pulley, the second back tensioning pulley, the second idler pulley, and the third back tensioning pulley are located on the slack side of the timing belt pulley. They not only support the timing belt pulley but also control the tension of the timing belt pulley by adjusting the position of the second idler pulley. The function of the inlet guide plate, outlet guide plate, and wire rope is to restrict the movement direction of the inner bag during cigarette pack transportation, ensuring that it moves along the required path. The wire rope is made of stainless steel wire with a diameter of 1 mm; The control compartment includes a vision-specific controller, a core control board, a power control module, and an emergency stop control module; A dedicated vision controller, equipped with a CPU and GPU, is configured to handle vision-related computations; The core control board includes a core control unit, a camera communication interface, a light source communication interface, input / output interfaces, and related circuits; The core control board communicates with the vision-dedicated controller via a serial port and is configured to be responsible for controlling the camera and light source. When the image sensor detects the internal card packet, the core control board lights up the light source and triggers the camera to take a picture. The core control board is responsible for communication with the electrical cabinet and control of the rejection valve. It controls the operation of the electromagnetic rejection valve in the rejection compartment based on the calculation results of the vision controller, receives the status information of the electrical cabinet, and outputs the statistical data to the electrical cabinet. The core control board is also responsible for controlling the servo motor, controlling the speed of the servo motor to exceed the running speed of the inner card pack conveyor belt, realizing the effective separation of the inner card pack, ensuring that each inner card pack is well isolated when it arrives at the shooting chamber, controlling the start and stop adjustment of the servo motor, and realizing linkage and synchronous start and stop with the entire packaging machine; The power control module includes a load disconnect switch, a filter, and a switching power supply; The load disconnect switch is the main power switch for the entire system and is configured to control the on / off of the external 220V power supply. The filter is configured to remove noise and spurious signals from the alternating current to ensure a stable and clean voltage for the system power supply. The switching power supply is configured to convert 220V power to 24V power to supply the power required by the vision-specific controller and the core control board. The emergency stop control module is configured to control the start and stop of the motor in emergency or abnormal situations. The imaging chamber includes six Ethernet industrial cameras, a ring LED light source, and an image sensor. The entire shooting chamber adopts a spherical integrating optical path structure, and its inner surface is covered with white matte paint; The six Ethernet industrial cameras are a top camera, a bottom camera, a front camera, a rear camera, a left-side camera, and a right-side camera; each Ethernet industrial camera is equipped with a low-distortion lens. The ring-shaped LED light source includes an upper light source and a lower light source. The ring-shaped LED light source is fixed in the middle of the sphere. The upper and lower light sources are arranged symmetrically. The upper and lower light sources do not directly illuminate the inner cardboard, but are diffusely reflected to the surface of the inner card pack through the inner surface of the shell, ensuring that all six sides of the entire inner card pack are uniformly illuminated. The shooting compartment has a circular hole the size of a camera lens at the center of the top, bottom, front, and back for taking pictures; a photo sensor port is located on the side of the camera hole at the top of the shooting compartment for the photo sensor to determine the position of the internal card holder; there is a rectangular hole at the center of the left and right sides for the internal card holder, belt, and limiting wire to pass through, and also for taking pictures. When the inner card bag reaches the designated position, the image sensor is triggered, and six sets of Ethernet industrial cameras capture images of six sides of the inner card bag and transmit them to the controller for processing. The rejection chamber includes an electromagnetic rejection valve, a booster nozzle, two rejection sensors, a rejection hood, and a rejection box. The booster nozzle and the rejection hood are located on both sides of the belt, with the booster nozzle aligned with the inlet of the rejection hood. The rejection box has a rectangular frame structure and is located below the rejection hood. It is used to collect defective inner card packs. The bottom of the rejection box is equipped with four rollers for pulling out. The two rejection sensors are symmetrically placed reflective sensors, located on both sides of the pressurized air nozzle. When the inner bag reaches the rejection position and both rejection sensors are blocked at the same time, if the controller's detection result is unqualified, the electromagnetic rejection air valve opens, and the pressurized air nozzle blows the defective inner bag into the rejection hood, and then it will slide into the rejection box. Human-machine interface, including touch LCD screen and rocker arm; The rocker arm adopts a three-axis structure, which enables it to be adjusted in three dimensions; the rocker arm is fixed to the side of the detection device with screws. Users browse system operation information and set various system parameters via the touchscreen LCD; the method includes the following steps: Step 1: Quick Start Phase; Initially, traditional vision methods were used to quickly detect and identify most common defects; Step 2: Collection and classification of sample data; The sample images of each camera are classified as qualified and unqualified. Unqualified samples are further classified into several categories of defects, including damage, missing parts, glue overflow, loose packaging, and inadequate bonding. Step 3: Training and enabling the deep learning model; For the collected positive and negative sample data from the 6 cameras, a deep learning classification model is trained for each camera. After obtaining the exclusive classification model for each camera, the deep learning detection function is enabled. Step 4: Iterative optimization of the model; Add invalid sample data and iterate the deep learning classification model for each camera; Step 3 specifically includes the following steps: Step 3.1: Model Building; Step 3.2: Dedicated model training; Step 3.3: Model deployment enabled; Step 4 specifically includes the following steps: Step 4.1: Model format conversion; Step 4.2: Import the dynamic link library; Step 4.3: Read in the category file; Step 4.4: Load the model file; Step 4.5: Convert the image into an input tensor; Step 4.6: Set the model input; Step 4.7: Perform network inference; Step 4.8: Obtain the reasoning results; Step 4.9: Iterative optimization of the model.

Images