A neural network-based chip character detection system

Abstract

The application is suitable for the field of industrial machine vision and automatic detection technology, and provides a chip character detection system based on a neural network, which comprises a continuous conveying device, a system controller and a cooperative imaging subsystem; the cooperative imaging subsystem comprises a plurality of vision imaging units; the system controller is configured to run an imaging scheduling and optimization neural network model; the cooperative imaging subsystem is configured to perform the following adaptive cooperative detection process, the system solves the problem that the imaging scheduling and optimization neural network model is used to divide the appropriate vision imaging units according to the conveying speed and the load in real time, supports two strategies of time-sharing trigger serial detection and follow-up synchronous detection, adapts to different production rhythms and detection requirements, and improves the flexibility and efficiency of the system; the system integrates fixed and mobile vision imaging units, can select the best imaging mode for different detection tasks, realizes omnidirectional and high-precision detection, and avoids the limitation of a single sensor.

Description

Technical Field

[0001] This invention relates to the field of industrial machine vision and automated inspection technology, and more specifically, to a chip character detection system based on neural networks. Background Technology

[0002] In the semiconductor manufacturing and packaging testing industry, characters on the chip surface (such as model numbers, batch numbers, and trademarks) are crucial for product traceability, quality control, and authenticity verification. Traditional chip character inspection mainly relies on manual visual inspection or fixed optical character recognition systems, which suffer from low inspection efficiency, poor consistency, susceptibility to ambient light interference, and difficulty in recognizing characters against complex backgrounds and low contrast. As chip packaging moves towards miniaturization and high integration, character sizes are further reduced, and printing processes are becoming more diverse, making it difficult for traditional inspection methods to meet the demands of high-precision, high-speed production.

[0003] Currently, there are automated inspection solutions on the market that use a single vision sensor combined with a fixed light source, but they typically lack adaptability to different imaging conditions (such as reflections, shadows, motion blur, etc.). Especially in continuously conveying production lines, chip position, orientation, and lighting conditions may change at any time, leading to unstable image quality and affecting the accuracy of character recognition. In addition, fixed shooting modes struggle to balance inspection speed and image quality, especially on high-speed production lines where missed or false detections are likely to occur.

[0004] Therefore, there is an urgent need for an intelligent inspection system that can adapt to complex production line environments and achieve dynamic imaging scheduling and adaptive optimization in order to improve the robustness, efficiency and accuracy of chip character detection. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a chip character detection system based on a neural network. By scheduling and optimizing the neural network model, the system allocates appropriate visual imaging units in real time according to the conveying speed and load, supports two strategies: time-sharing triggering serial detection and follow-up synchronous detection, adapts to different production rhythms and detection requirements, and improves system flexibility and efficiency.

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

[0007] A chip character detection system based on neural networks includes a continuous conveying device, a system controller, and a collaborative imaging subsystem;

[0008] The collaborative imaging subsystem includes multiple visual imaging units;

[0009] The system controller is configured to run an imaging scheduling and optimization neural network model;

[0010] The collaborative imaging subsystem is configured to perform the following adaptive collaborative detection process:

[0011] Imaging scheduling phase: Based on the speed and load of the continuous conveying device, the system controller dynamically allocates at least one dedicated visual imaging unit to the chip trays flowing through different physical sections through imaging scheduling and optimization neural network models, and generates an initial imaging scheduling strategy.

[0012] The initial imaging scheduling strategy is one of the following detection strategies or a combination thereof:

[0013] Time-division triggering serial detection strategy: The visual imaging unit is a fixed visual imaging unit; the fixed visual imaging unit is controlled to sequentially capture images of the same chip tray at different time points;

[0014] Follow-up synchronous detection strategy: includes a synchronization unit; the visual imaging unit is a mobile visual imaging unit; the control synchronization unit drives at least one mobile visual imaging unit to move synchronously with the corresponding chip tray, and completes the shooting during the movement;

[0015] Adaptive optimization phase: Acquire chip images from the visual imaging unit during the imaging scheduling phase; input the chip images into the imaging scheduling and optimization neural network model, which analyzes the image quality and identifies the imaging challenge of chip characters; based on the imaging challenge, the imaging scheduling and optimization neural network model dynamically generates optimization instructions;

[0016] Collaborative decision-making stage: Based on the final image obtained after the adaptive optimization stage, the system completes the final detection and judgment of the chip characters.

[0017] The present invention is further configured such that the optimization instructions include at least one of the following:

[0018] Adjust the illumination parameters of the current or subsequent visual imaging units;

[0019] Adjust the camera parameters of the current or subsequent visual imaging units;

[0020] The initial imaging scheduling strategy is modified or regenerated.

[0021] The invention is further configured such that: the continuous conveying device includes a conveying frame; four sets of rotating shafts are symmetrically and rotatably arranged through the inner walls of the conveying frame; encoders are installed on the rotating shafts; pulleys are symmetrically fixed on the circumferential sides of the rotating shafts; a matching belt is provided between two opposite pulleys for transmission; a plurality of receiving grooves are evenly opened on the outer wall of the belt; two coaxial receiving grooves form a receiving cavity; a servo motor is fixedly installed on the side of the conveying frame; the output end of the servo motor is fixedly connected to the end of a rotating shaft; and a first gear that meshes with each other is fixed on the circumferential sides of two adjacent rotating shafts.

[0022] The invention is further configured such that: a support platform is provided through the inside of the conveyor frame; multiple chip trays are slidably disposed on the surface of the support platform; and connecting columns adapted to the corresponding receiving cavities are symmetrically fixed on opposite sides of the chip trays.

[0023] The present invention is further configured such that: the fixed visual imaging unit includes a top plate fixedly installed on the top of the conveyor; three sets of first industrial cameras are fixedly installed on the bottom surface of the top plate in sequence; a first lens is installed on the first industrial camera; and a coaxial light source for character detection, a dome light source for appearance detection, and a low-angle light source for pin detection are respectively installed at the bottom of the three sets of first industrial cameras.

[0024] The invention is further configured such that: the synchronization unit includes a drive shaft symmetrically and rotatably mounted on the conveyor frame; a conveyor roller is fixed to the circumferential side of the drive shaft; a conveyor belt adapted to the two conveyor rollers is driven between them; a second gear is fixed to the end of a set of drive shafts; and a third gear that meshes with the second gear is fixed to the end of a set of rotating shafts.

[0025] The present invention is further configured such that: the mobile vision imaging unit includes a mounting plate fixed to the surface of the conveyor belt; an inner light-shielding cylinder is fixed to the surface of the mounting plate; a second industrial camera is mounted on the surface of the mounting plate; a second lens is mounted on the second industrial camera; a pulse light source is installed inside the inner light-shielding cylinder; guide rods are symmetrically fixed to the surface of the mounting plate; and an outer light-shielding cylinder is slidably arranged between the two sets of guide rods.

[0026] The invention is further configured such that: a sliding rod is fixed to the outer wall of the outer light-shielding cylinder; an annular track with sliding cooperation of the sliding rod is fixedly installed on the conveyor frame; inclined tracks are provided at both ends of the bottom of the annular track and are interconnected; a straight track is connected between the two sets of inclined tracks; and pressure switches for controlling the start and stop of the second industrial camera are fixedly installed on the surface of the straight track near both ends.

[0027] The advantages of this invention are:

[0028] This invention uses imaging scheduling and an optimized neural network model to allocate appropriate visual imaging units in real time according to the conveying speed and load. It supports two strategies: time-sharing triggered serial detection and follow-up synchronous detection, adapting to different production rhythms and detection requirements, and improving system flexibility and efficiency.

[0029] This invention is based on real-time acquired images. It uses neural networks to analyze imaging challenges (such as uneven lighting, motion blur, and reflection interference) and dynamically adjusts the light source, camera parameters, or scheduling strategies to improve image quality from the source and enhance the robustness of character recognition.

[0030] This invention integrates two types of visual imaging units: fixed and mobile. It can select the best imaging method for different detection tasks (characters, appearance, pins) to achieve all-round, high-precision detection and avoid the limitations of a single sensor.

[0031] The continuous conveying device and imaging unit of this invention achieve continuous conveying and synchronous imaging of chip trays through mechanical and control coordination, supporting non-stop inspection and significantly improving production line throughput. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the structure of a chip character detection system based on a neural network according to the present invention.

[0033] Figure 2 This is a flowchart illustrating the adaptive collaborative detection process of the present invention.

[0034] Figure 3 This is a schematic diagram of the continuous conveying device of the present invention.

[0035] Figure 4 For the present invention Figure 3 A structural diagram from a frontal viewpoint.

[0036] Figure 5 This is a schematic diagram of the structure of the fixed visual imaging unit of the present invention.

[0037] Figure 6 This is a schematic diagram of the detection structure in Embodiment 2 of the present invention.

[0038] Figure 7 For the present invention Figure 6 A perspective that faces the issue directly.

[0039] Figure 8 For the present invention Figure 7 Enlarged view of region A.

[0040] Figure 9 This is a schematic diagram of the detection structure in Embodiment 3 of the present invention.

[0041] Figure 10 For the present invention Figure 9 Enlarged view of region B.

[0042] In the diagram: 1. Continuous conveying device; 2. Chip tray; 3. Synchronization unit; 4. Conveyor frame; 5. Rotary shaft; 6. Pulley; 7. Belt; 8. Receiving groove; 9. Receiving cavity; 10. Servo motor; 11. Sliding rod; 12. Circular track; 13. Support platform; 14. Connecting column; 15. Top plate; 16. First industrial camera; 17. First lens; 18. Coaxial light source; 19. Dome light source; 20. Low-angle light source; 21. Drive shaft; 22. Conveyor roller; 23. Conveyor belt; 24. Inclined track; 25. Straight track; 26. Pressure switch; 27. Mounting plate; 28. Inner light shield; 29. ​​Second industrial camera; 30. Second lens; 31. Pulse light source; 32. Guide rod; 33. Outer light shield. Detailed Implementation

[0043] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0044] It should be noted that, unless otherwise specified, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0045] In this invention, unless otherwise stated, the directional terms such as "up" and "down" generally refer to the directions shown in the accompanying drawings, or to the vertical, perpendicular, or gravitational direction; similarly, for ease of understanding and description, "left" and "right" generally refer to the left and right shown in the accompanying drawings; "inner" and "outer" refer to the inner and outer contours of each component itself, but the above directional terms are not intended to limit this invention.

[0046] Example 1, please refer to Figure 1-10 The present invention provides the following technical solutions:

[0047] A chip character detection system based on a neural network, specifically, includes a continuous conveying device 1, a system controller, and a collaborative imaging subsystem; the collaborative imaging subsystem includes multiple visual imaging units;

[0048] The system controller is configured to run an imaging scheduling and optimization neural network model;

[0049] The collaborative imaging subsystem is configured to perform the following adaptive collaborative detection process:

[0050] Imaging scheduling stage: Based on the speed and load of the continuous conveying device 1, the system controller dynamically allocates at least one dedicated visual imaging unit to the chip trays 2 flowing through different physical sections through imaging scheduling and optimization neural network model, and generates an initial imaging scheduling strategy.

[0051] The initial imaging scheduling strategy is one of the following detection strategies or a combination thereof:

[0052] Time-division triggering serial detection strategy: The visual imaging unit is a fixed visual imaging unit; the fixed visual imaging unit is controlled to sequentially capture images of the same chip tray 2 at different time points;

[0053] Follow-up synchronous detection strategy: including synchronization unit 3; the visual imaging unit is a mobile visual imaging unit; control synchronization unit 3 to drive at least one mobile visual imaging unit to move synchronously with the corresponding chip tray 2, and complete the shooting during the movement;

[0054] Adaptive optimization phase: Acquire chip images from the visual imaging unit during the imaging scheduling phase; input the chip images into the imaging scheduling and optimization neural network model, which analyzes the image quality and identifies the imaging challenge of chip characters; based on the imaging challenge, the imaging scheduling and optimization neural network model dynamically generates optimization instructions;

[0055] Collaborative decision-making stage: Based on the final image obtained after the adaptive optimization stage, the system completes the final detection and judgment of the chip characters.

[0056] The optimization instructions include at least one of the following:

[0057] Adjust the illumination parameters of the current or subsequent visual imaging units;

[0058] Adjust the camera parameters of the current or subsequent visual imaging units;

[0059] The initial imaging scheduling strategy is modified or regenerated.

[0060] The continuous conveying device 1 includes a conveying frame 4; four sets of rotating shafts 5 are symmetrically and rotatably arranged through the inner walls of the conveying frame 4; pulleys 6 are symmetrically fixed on the sides of the rotating shafts 5; a matching belt 7 is provided between two opposite pulleys 6 for transmission; several receiving grooves 8 are evenly opened on the outer wall of the belt 7; two coaxial receiving grooves 8 form a receiving cavity 9; a servo motor 10 is fixedly installed on the side of the conveying frame 4; the output end of the servo motor 10 is fixedly connected to the end of a rotating shaft 5; and a first gear that meshes with each other is fixed on the sides of two adjacent rotating shafts 5.

[0061] A support platform 13 is installed inside the conveyor frame 4; multiple chip trays 2 are slidably installed on the surface of the support platform 13; the chip trays 2 are symmetrically fixed with connecting columns 14 that are adapted to the corresponding receiving cavities 9 on their opposite sides.

[0062] Working principle of this embodiment:

[0063] By placing multiple chip trays 2 sequentially onto the support platform 13 and pushing the outermost chip tray 2, each chip tray 2 moves towards the belt 7. The servo motor 10 is controlled to drive the corresponding shaft 5 to rotate, which in turn drives the two meshing first gears to rotate, thereby driving the lower shaft 5 to rotate synchronously in the opposite direction. This, in turn, drives the two sets of belts 7 arranged oppositely to rotate synchronously in the opposite direction. The connecting posts 14 at both ends of the chip tray 2 closest to the belt 7 are engaged into the corresponding receiving cavity 9, thus completing the sequential and continuous conveying of multiple chip trays 2.

[0064] The imaging scheduling and optimization neural network model is a multi-task deep learning model deployed on embedded systems or industrial control computers. It possesses dual capabilities: dynamic scheduling decision-making and image quality optimization analysis. The model employs an encoder-multi-branch decoder structure, specifically including:

[0065] Feature extraction encoder: It uses convolutional neural networks (such as ResNet, EfficientNet, etc.) to perform multi-level feature extraction on the input chip image and outputs a feature map containing global and local information.

[0066] Scheduling decision branch: Based on the characteristics of the encoder output, combined with the real-time status information of the conveying device from the encoder (such as speed, position, load), through the fully connected layer and attention mechanism, output the type of visual imaging unit (fixed / mobile) and its scheduling strategy (time-sharing triggering / follow-up synchronization) that should be allocated to the current chip tray 2.

[0067] Imaging Analysis Branch: Performs image quality assessment and imaging challenge identification on the same feature map, with outputs including but not limited to:

[0068] Image blur rating; illumination uniformity analysis; reflective area detection; character region contrast evaluation; motion artifact recognition.

[0069] Model training and optimization

[0070] The neural network is trained using a combination of end-to-end supervised learning and reinforcement learning:

[0071] Supervised learning phase: Using a labeled chip image dataset, the model is trained to identify image quality problems and imaging challenges, and learns to generate optimization instructions (such as light source parameter adjustment and camera parameter suggestions) under different imaging conditions.

[0072] Reinforcement learning phase: In the simulation environment, the model learns how to make optimal scheduling decisions in a dynamic production line environment by interacting with virtual conveyor devices and imaging units to maximize detection success rate and throughput.

[0073] Cooperative mechanism with character recognition neural network

[0074] The system can integrate a dedicated character recognition neural network (such as CRNN or Transformer-based OCR model) for character detection and recognition in the optimized final image. This model can be deployed in series or in parallel with the imaging scheduling model, supporting the following two working modes:

[0075] Serial mode: After the imaging scheduling model outputs an optimized image, it is directly input into the character recognition network for final judgment.

[0076] Parallel collaborative mode: The two models share the feature extraction layer, enabling simultaneous imaging optimization and character recognition, further improving the system's real-time performance.

[0077] Real-time inference and deployment

[0078] The system controller deploys a lightweight inference engine (such as TensorRT or OpenVINO) to support the real-time operation of the neural network on embedded devices or industrial control computers. The model supports online learning and incremental updates, and can continuously optimize scheduling and imaging strategies based on actual production line data.

[0079] By using imaging scheduling and optimizing neural network models, appropriate visual imaging units are allocated in real time according to conveying speed and load. It supports two strategies: time-sharing triggered serial detection and follow-up synchronous detection, adapting to different production rhythms and detection needs, and improving system flexibility and efficiency. Based on real-time acquired images, the neural network analyzes imaging challenges (such as uneven illumination, motion blur, and reflection interference) and dynamically adjusts light source, camera parameters, or scheduling strategies to improve image quality from the source and enhance character recognition robustness. The system integrates both fixed and mobile visual imaging units, allowing for the selection of the optimal imaging method for different detection tasks (characters, appearance, pins), achieving all-round, high-precision detection and avoiding the limitations of a single sensor.

[0080] Example 2, please refer to Figure 1-5 This second embodiment is an improvement on the first embodiment as follows: Specifically, an encoder is installed on the rotating shaft 5; the fixed vision imaging unit includes a top plate 15 fixedly installed on the top of the conveyor frame 4; three sets of first industrial cameras 16 are fixedly installed on the bottom surface of the top plate 15 in sequence; a first lens 17 is installed on the first industrial camera 16; and a coaxial light source 18 for character detection, a dome light source 19 for appearance detection, and a low-angle light source 20 for pin detection are respectively installed at the bottom of the three sets of first industrial cameras 16.

[0081] Working principle of this embodiment two:

[0082] The coaxial light source 18 is designed to detect tiny bumps or engravings (such as chip characters) on a smooth surface, and its core feature is that the optical path is parallel to the camera's optical axis.

[0083] Core component: Beam splitter, a semi-transparent and semi-reflective mirror placed at a 45-degree angle in front of the camera lens; LED light source enters horizontally from the side; after passing through the beam splitter, the light is reflected vertically downwards and illuminates the product surface, and the light reflected vertically back from the surface passes through the beam splitter again and enters the camera lens directly.

[0084] It is a standalone component, usually fixed directly to the front of the camera lens via a connector or mounting bracket. The entire "camera-lens-coaxial light source" is mounted on the bracket as a single module.

[0085] Chip detection function: A smooth chip surface reflects light vertically back to the camera, appearing bright white; a rough, concave or raised character will cause diffuse reflection, with only a small amount of light returning to the camera, appearing dark black.

[0086] Effect: Creates a high-contrast black and white image, making it the preferred light source for character reading.

[0087] The Dome Light Source 19 is designed to eliminate reflections and shadows, providing shadowless, uniform illumination.

[0088] Installation structure:

[0089] A hemispherical dome with an inner wall made of diffuse reflective material.

[0090] LED light source: Usually installed on the side or top of the dome, the light shines on the inner wall of the hemisphere.

[0091] Working principle: After multiple diffuse reflections on the inner wall of the hemisphere, the light shines evenly and softly onto the object being tested from all angles.

[0092] The dome light source 19 is a separate and large fixed device. The industrial camera takes pictures from inside the dome through the opening at the top of the dome. The dome light is fixed on the frame and completely covers the chip tray 2 that needs to be inspected inside it.

[0093] Chip testing function:

[0094] Eliminate glare: It can greatly reduce the specular reflections generated by the smooth surface and pins of the chip.

[0095] Eliminate shadows: Achieve shadowless illumination so that every detail on the chip surface is evenly illuminated.

[0096] Results: Obtains a uniform image with clear details, which is very suitable for overall appearance inspection and color recognition.

[0097] Low-angle light source 20 is used to highlight the three-dimensional contours and height differences of objects.

[0098] Installation structure:

[0099] Long strip-shaped LED light bead array.

[0100] During installation, light passes across the surface of the object being measured at a very small angle (almost parallel).

[0101] Installation method:

[0102] Installed directly around the camera lens: used as a low-angle ring light, with light emanating from around the lens at a low angle.

[0103] Independent installation: One or more strip lights are individually fixed on the frames on both sides of the camera, and their illumination angle is precisely adjusted so that they are almost parallel to the chip surface.

[0104] Dark field lighting: This is an extreme form of low-angle lighting, with an even lower installation angle, so that light reflected from a flat surface cannot enter the lens at all (appearing black), while only the uneven edges can scatter light into the lens (appearing white).

[0105] Chip testing function:

[0106] Highlighting pins: The pins of a chip can block light, creating a bright edge on the other side, while the bottom of the pins or the location where a pin is missing will create a shadow.

[0107] Detecting scratches / bumps: Any minute surface undulations will produce noticeable differences in light and shadow when illuminated.

[0108] Effect: Converts height information into grayscale information, mainly used for pin detection, warpage detection, scratch detection, etc.

[0109] A high-quality encoder is installed on the rotating shaft 5 to generate a unified position clock signal. When the first industrial camera 16 takes an image at time T, the system controller needs to accurately calculate which part of the chip tray 2 the image belongs to based on the encoder position and the conveyor belt speed, and correctly associate the image data with the detection task of the chip tray 2, so as to realize "non-stop detection" of the chips inside the chip tray 2, thereby improving detection efficiency and effect.

[0110] Example 3, please refer to Figure 1 , Figure 2 , Figure 6-10 This embodiment three is an improvement on the first embodiment as follows: Specifically, the synchronization unit 3 includes a drive shaft 21 symmetrically and rotatably mounted on the conveyor frame 4; a conveyor roller 22 is fixed on the circumferential side of the drive shaft 21; a conveyor belt 23 adapted to the two conveyor rollers 22 is driven between them; a second gear is fixed at the end of a set of drive shafts 21; and a third gear that meshes with the second gear is fixed at the end of a set of rotating shafts 5.

[0111] The mobile vision imaging unit includes a mounting plate 27 fixed to the surface of the conveyor belt 23; an inner light-shielding cylinder 28 fixed to the surface of the mounting plate 27; a second industrial camera 29 mounted on the surface of the mounting plate 27; a second lens 30 mounted on the second industrial camera 29; a pulse light source 31 installed inside the inner light-shielding cylinder 28; guide rods 32 symmetrically fixed to the surface of the mounting plate 27; and an outer light-shielding cylinder 33 slidably disposed between the two sets of guide rods 32.

[0112] A sliding rod 11 is fixed to the outer wall of the outer light shield 33; a ring track 12 with sliding cooperation of the sliding rod 11 is fixedly installed on the conveyor frame 4; inclined tracks 24 are connected to each other at both ends of the bottom of the ring track 12; a straight track 25 is connected between the two sets of inclined tracks 24; pressure switches 26 for controlling the start and stop of the second industrial camera 29 are fixedly installed on the surface of the straight track 25 near both ends.

[0113] Working principle of this embodiment three:

[0114] By controlling the start servo motor 10 to drive the corresponding upper and lower opposing belts 7 to rotate synchronously in opposite directions, and through the meshing of the second gear and the third gear, the synchronous transmission of the conveyor belt 23 is achieved, thereby driving each group of mobile vision imaging units and the chip tray 2 to drive synchronously. That is, each group of mobile vision imaging units can drive synchronously with the corresponding chip tray 2 without waiting, thus improving the detection efficiency.

[0115] When a set of mobile vision imaging units slides from the circular track 12 to the inclined track 24, the slide bar 11 is guided by the inclined track 24, causing it to descend during the movement. This drives the outer light-shielding tube 33 to slide down along the guide rod 32. When the corresponding mobile vision imaging unit slides from the inclined track 24 to the straight track 25, the outer light-shielding tube 33 covers the corresponding chip tray 2, effectively isolating ambient light and interference from adjacent workstations, ensuring a stable imaging environment, improving detection consistency, and enhancing detection quality. When the connecting post 14 on the mobile vision imaging unit presses the first set of pressure switches 26, the pressure switch 26 transmits a signal to the system controller. The system controller then controls the corresponding second industrial camera 29 and pulse light source 31 to start, enabling the outer light-shielding tube 33 to cover the corresponding chip tray 2 for detection. When the corresponding connecting post 14 presses the second set of pressure switches 26, the system controller controls the corresponding second industrial camera 29 and pulse light source 31 to shut down.

[0116] Obviously, the embodiments described above are merely some, not all, embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0117] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0118] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0119] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0120] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A chip character detection system based on a neural network, characterized in that: Includes a continuous transport device, a system controller, and a collaborative imaging subsystem; The collaborative imaging subsystem includes multiple visual imaging units; The system controller is configured to run an imaging scheduling and optimization neural network model; The collaborative imaging subsystem is configured to perform the following adaptive collaborative detection process: Imaging scheduling phase: Based on the speed and load of the continuous conveying device, the system controller dynamically allocates at least one dedicated visual imaging unit to the chip trays flowing through different physical sections through imaging scheduling and optimization neural network models, and generates an initial imaging scheduling strategy. The initial imaging scheduling strategy is a combination of the following detection strategies: Time-division triggering serial detection strategy: The visual imaging unit is a fixed visual imaging unit; the fixed visual imaging unit is controlled to sequentially capture images of the same chip tray at different time points; Follow-up synchronous detection strategy: includes a synchronization unit; the visual imaging unit is a mobile visual imaging unit; the control synchronization unit drives at least one mobile visual imaging unit to move synchronously with the corresponding chip tray, and completes the shooting during the movement; Adaptive optimization phase: Acquire chip images from the visual imaging unit during the imaging scheduling phase; input the chip images into the imaging scheduling and optimization neural network model, which analyzes the image quality and identifies the imaging challenge of chip characters; Based on imaging challenges, an imaging scheduling and optimization neural network model dynamically generates optimization instructions. Collaborative decision-making stage: Based on the final image obtained after the adaptive optimization stage, the system completes the final detection and judgment of the chip characters; The continuous conveying device includes a conveyor frame; the synchronization unit includes a drive shaft symmetrically and rotatably mounted on the conveyor frame; conveyor rollers are fixed to the circumferential side of the drive shaft; and a conveyor belt adapted to the two conveyor rollers is driven between them. The mobile vision imaging unit includes a mounting plate fixed to the surface of a conveyor belt; an inner light-shielding cylinder fixed to the surface of the mounting plate; a second industrial camera mounted on the surface of the mounting plate; a second lens mounted on the second industrial camera; a pulse light source installed inside the inner light-shielding cylinder; guide rods symmetrically fixed to the surface of the mounting plate; and an outer light-shielding cylinder slidably disposed between the two sets of guide rods. A sliding rod is fixed to the outer wall of the outer light-shielding cylinder; a ring track with sliding cooperation of the sliding rod is fixedly installed on the conveyor frame; inclined tracks are connected to each other at both ends of the bottom of the ring track; straight tracks are connected between the two sets of inclined tracks; pressure switches for controlling the start and stop of the second industrial camera are fixedly installed on the surface of the straight track near both ends.

2. The chip character detection system based on a neural network according to claim 1, characterized in that: The optimization instructions include at least one of the following: Adjust the illumination parameters of the current or subsequent visual imaging units; Adjust the camera parameters of the current or subsequent visual imaging units; The initial imaging scheduling strategy is modified or regenerated.

3. The chip character detection system based on a neural network according to claim 2, characterized in that: Four sets of rotating shafts are symmetrically arranged through the inner wall of the conveyor frame; encoders are installed on the rotating shafts; pulleys are symmetrically fixed on the circumference of the rotating shafts; a matching belt is provided between two opposite pulleys for transmission; several receiving grooves are evenly opened on the outer wall of the belt; two receiving grooves on the same axis form a receiving cavity; a servo motor is fixedly installed on the side of the conveyor frame; the output end of the servo motor is fixedly connected to the end of a rotating shaft; a first gear that meshes with each other is fixed on the circumference of two adjacent rotating shafts.

4. The chip character detection system based on a neural network according to claim 3, characterized in that: A support platform is installed inside the conveyor frame; multiple chip trays are slidably mounted on the surface of the support platform; the chip trays are symmetrically fixed with connecting columns that are adapted to the corresponding receiving cavities on their opposite sides.

5. A chip character detection system based on a neural network according to claim 4, characterized in that: The fixed vision imaging unit includes a top plate fixedly installed on the top of the conveyor; three sets of first industrial cameras are fixedly installed on the bottom surface of the top plate in sequence; a first lens is installed on the first industrial camera; and a coaxial light source for character detection, a dome light source for appearance detection, and a low-angle light source for pin detection are respectively installed at the bottom of the three sets of first industrial cameras.

6. The chip character detection system based on a neural network according to claim 4, characterized in that: A second gear is fixed to the end of a set of drive shafts; a third gear that meshes with the second gear is fixed to the end of a set of rotating shafts.

Images