End face image feature enhancement segmentation method based on deep convolutional network

Abstract

The invention discloses an end face image feature enhancement segmentation method based on a deep convolutional network, belongs to the technical field of image feature enhancement, and is used for end face image feature enhancement. Comprising the steps that photoelectricity is triggered when equipment runs, hard triggering is conducted through a camera, and varieties, yarn colors and batches and surface defects are detected respectively; the camera AI model is divided into three threads based on a YOLOv8 deep learning model to process camera data; and summarizing structured data processed by the camera AI model to machine learning, pushing qualified structured data into a normal channel, and rejecting unqualified structured data. Yarn foreign fibers are detected by purple light, varieties are detected by a variety camera, and surface defects are detected by a quality camera. The cone yarn sorting efficiency is improved, manpower in the whole packaging process is reduced, the product accuracy is improved, the omission ratio is reduced, and zero batch mixing is achieved; and the productivity and the product percent of pass are improved.

Description

Technical Field

[0001] This invention discloses an end-face image feature enhancement and segmentation method based on deep convolutional networks, belonging to the field of image feature enhancement technology. Background Technology

[0002] In recent years, with the deep penetration of intelligent manufacturing technology into the textile industry, yarn package visual recognition and online sorting systems have become a key breakthrough for achieving less manpower or even unmanned operation in packaging processes. Traditional rule-based methods rely on manually setting thresholds and template matching, which suffers from bottlenecks such as poor generalization ability, high maintenance costs, and difficulty in compressing cycle time in multi-task parallel scenarios such as variety identification, color verification, and defect detection. The rise of deep convolutional networks has provided a new approach to these problems: through end-to-end feature learning, algorithms can maintain high accuracy under conditions such as complex backgrounds, lighting fluctuations, and high-speed motion, and have therefore been rapidly introduced into the textile quality inspection process.

[0003] However, most existing studies focus on single-task models. Faced with the actual needs of yarn production lines where three types of targets coexist—variety, foreign fibers, and yarn surface defects—and the wide range of detection scales, the following challenges still need to be addressed: redundant multi-task network structures and serious waste of computing resources; high rate of missed detection of small sample defects under high-speed production; and limited GPU memory in industrial settings, making it difficult to balance accuracy and real-time performance. Summary of the Invention

[0004] The purpose of this invention is to provide an end-face image feature enhancement and segmentation method based on deep convolutional networks to solve the problems in the prior art, such as the high dependence on manual labor in yarn packaging, low sorting efficiency, high risk of batch mixing and missorting, gaps in defect traceability, rigid packaging mode, and rising labor costs.

[0005] A face image feature enhancement and segmentation method based on deep convolutional networks, comprising: S1. When the equipment is running, photoelectric sensors are triggered, and hard triggering is performed through a camera to detect the variety, yarn foreign fibers, and surface defects respectively. S2. The camera AI model is based on the YOLOv8 deep learning model and is divided into three threads to process camera data. The camera light source is controlled alternately, and exposure and shooting are performed. S3. The structured data processed by the camera AI model is aggregated into the machine learning system. Qualified data is pushed into the normal channel, while unqualified data is discarded.

[0006] Ultraviolet light is used to detect foreign fibers in yarn, and a camera is used to detect the variety and surface defects. The variety is determined by the detected yarn color and pattern.

[0007] Triggering modes include soft triggering and hard triggering; Soft triggering includes commanding the software system to control the camera or software controlling the light source to match and take pictures; Hard triggering is automatically initiated by the equipment to carry out normal production.

[0008] S2 includes camera AI models, including the violet depth AI model, the variety depth AI model, and the quality depth AI model.

[0009] The Ziguang Deep AI model detects foreign fibers in yarn, the Variety Deep AI machine detects the variety, and the Quality Deep AI model queries for defects such as fuzz, stains, and broken yarns.

[0010] The YOLOv8 deep learning model consists of a backbone network, a connectivity network, and a task head.

[0011] The backbone network consists of convolutional kernels, alternating modules, and fast spatial pyramid layers. The alternating modules include four sets of cascaded fine-grained convolutional modules and C2f modules. The four C2f modules contain bottleneck models of 3, 6, 6, and 3, respectively.

[0012] The connection part includes three cascaded upward processing models and three cascaded downward processing modules. The upward processing module includes an upsampling layer, a feature fusion layer, and a C2f module. The downward processing module includes a C2f module, a fine-grained convolution module, and a feature fusion layer. The task header consists of four convolutional layers, and the four C2f modules of the down-processing module are connected to the four convolutional layers respectively.

[0013] The second C2f module of the backbone network is connected to the second feature fusion layer of the upward processing module, and the third C2f module of the backbone network is connected to the first feature fusion layer of the upward processing module.

[0014] Compared with existing technologies, the present invention has the following advantages: it improves the efficiency of yarn bobbin sorting, reduces manual labor in the entire packaging process, improves product accuracy, reduces the rate of missed inspections, and achieves zero mixed batches; it also increases production capacity and product qualification rate. Attached Figure Description

[0015] Figure 1 This is the overall flowchart of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0017] A method for enhancing and segmenting end-face images based on deep convolutional networks, the overall flowchart of which is shown below. Figure 1 As shown, it includes: S1. When the equipment is running, photoelectric sensors are triggered, and hard triggering is performed through a camera to detect the variety, yarn foreign fibers, and surface defects respectively. S2. The camera AI model is based on the YOLOv8 deep learning model and is divided into three threads to process camera data. The camera light source is controlled alternately, and exposure and shooting are performed. S3. The structured data processed by the camera AI model is aggregated into the machine learning system. Qualified data is pushed into the normal channel, while unqualified data is discarded.

[0018] Ultraviolet light is used to detect foreign fibers in yarn, and a camera is used to detect the variety and surface defects. The variety is determined by the detected yarn color and pattern.

[0019] Triggering modes include soft triggering and hard triggering; Soft triggering includes commanding the software system to control the camera or software controlling the light source to match and take pictures; Hard triggering is automatically initiated by the equipment to carry out normal production.

[0020] S2 includes camera AI models, including the violet depth AI model, the variety depth AI model, and the quality depth AI model.

[0021] The Ziguang Deep AI model detects foreign fibers in yarn, the Variety Deep AI machine detects the variety, and the Quality Deep AI model queries for defects such as fuzz, stains, and broken yarns.

[0022] The YOLOv8 deep learning model consists of a backbone network, a connectivity network, and a task head.

[0023] The backbone network consists of convolutional kernels, alternating modules, and fast spatial pyramid layers. The alternating modules include four sets of cascaded fine-grained convolutional modules and C2f modules. The four C2f modules contain bottleneck models of 3, 6, 6, and 3, respectively.

[0024] The connection part includes three cascaded upward processing models and three cascaded downward processing modules. The upward processing module includes an upsampling layer, a feature fusion layer, and a C2f module. The downward processing module includes a C2f module, a fine-grained convolution module, and a feature fusion layer. The task header consists of four convolutional layers, and the four C2f modules of the down-processing module are connected to the four convolutional layers respectively.

[0025] The second C2f module of the backbone network is connected to the second feature fusion layer of the upward processing module, and the third C2f module of the backbone network is connected to the first feature fusion layer of the upward processing module.

[0026] The camera uses a high-resolution industrial camera (≥5 megapixels) + LED light source, deployed at the top of the sorting station to achieve 360° blind-spot-free image acquisition. One camera controls the brightness of the light source by controlling the LED white light / UV light, and alternately controls the camera to take pictures accordingly, achieving a single yarn cycle of 800ms for UV light / variety / quality.

[0027] Based on the YOLOv8 deep learning model, the training set covers 100,000 defect samples of 12 categories, with an identification accuracy of ≥98%, and supports self-learning of paper tube color varieties (the model iteration is completed in ≤30 minutes when adding new varieties).

[0028] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for enhancing and segmenting end-face images based on deep convolutional networks, characterized in that, include: S1. When the equipment is running, photoelectric sensors are triggered, and hard triggering is performed through a camera to detect the variety, yarn foreign fibers, and surface defects respectively. S2. The camera AI model is based on the YOLOv8 deep learning model and is divided into three threads to process camera data. The camera light source is controlled alternately, and exposure and shooting are performed. S3. The structured data processed by the camera AI model is aggregated into the machine learning system. Qualified data is pushed into the normal channel, while unqualified data is discarded.

2. The end-face image feature enhancement and segmentation method based on deep convolutional networks according to claim 1, characterized in that, Ultraviolet light is used to detect foreign fibers in yarn, and a camera is used to detect the variety and surface defects. The variety is determined by the detected yarn color and pattern.

3. The end-face image feature enhancement and segmentation method based on deep convolutional networks according to claim 2, characterized in that, Triggering modes include soft triggering and hard triggering; Soft triggering includes commanding the software system to control the camera or software controlling the light source to match and take pictures; Hard triggering is automatically initiated by the equipment to carry out normal production.

4. The end-face image feature enhancement and segmentation method based on deep convolutional networks according to claim 3, characterized in that, S2 includes camera AI models, including the violet depth AI model, the variety depth AI model, and the quality depth AI model.

5. The end-face image feature enhancement and segmentation method based on deep convolutional networks according to claim 4, characterized in that, The Ziguang Deep AI model detects foreign fibers in yarn, the Variety Deep AI machine detects the variety, and the Quality Deep AI model queries for defects such as fuzz, stains, and broken yarns.

6. The end-face image feature enhancement and segmentation method based on deep convolutional networks according to claim 5, characterized in that, The YOLOv8 deep learning model consists of a backbone network, a connectivity network, and a task head.

7. The end-face image feature enhancement and segmentation method based on deep convolutional networks according to claim 6, characterized in that, The backbone network consists of convolutional kernels, alternating modules, and fast spatial pyramid layers. The alternating modules include four sets of cascaded fine-grained convolutional modules and C2f modules. The four C2f modules contain bottleneck models of 3, 6, 6, and 3, respectively.

8. The end-face image feature enhancement and segmentation method based on deep convolutional networks according to claim 7, characterized in that, The connection part includes three cascaded upward processing models and three cascaded downward processing modules. The upward processing module includes an upsampling layer, a feature fusion layer, and a C2f module. The downward processing module includes a C2f module, a fine-grained convolution module, and a feature fusion layer. The task header consists of four convolutional layers, and the four C2f modules of the down-processing module are connected to the four convolutional layers respectively.

9. The end-face image feature enhancement and segmentation method based on deep convolutional networks according to claim 8, characterized in that, The second C2f module of the backbone network is connected to the second feature fusion layer of the upward processing module, and the third C2f module of the backbone network is connected to the first feature fusion layer of the upward processing module.

Images