Real-time Surface Defect Detection of Synthetic Fiber Filament Bobbins Using Lightweight YOLOv8 for Automated Textile Quality Inspection

This repository contains the source code and partial benchmark dataset associated with the paper entitled Real-time Surface Defect Detection of Synthetic Fiber Filament Bobbins Using Lightweight YOLOv8 for Automated Textile Quality Inspection. This repository contains the source code and partial benchmark dataset associated with the paper entitled Real-Time Application of Industrial-Grade Lightweight Algorithm for Surface Defect Detection in Chemical Fiber Yarn Cakes. This code repository contains the source code and some benchmark datasets related to the paper titled "Real-time Application of an Industrial-grade Lightweight Algorithm for Surface Defect Detection of Chemical Fiber Yarn Rolls". This work focuses on industrial deployment and edge deployment of surface defect detection for chemical fiber yarn cakes. We propose a highly simplified model quantization mechanism, a novel model construction strategy, and an improved training pipeline. The framework includes automated detection head search, structure recombination, model update, as well as model pruning and quantization, achieving efficient and real-time inference on edge devices.This work focuses on industrial deployment and edge deployment of surface defect detection for chemical fiber yarn cakes. We propose a highly simplified model quantization mechanism, a novel model construction strategy, and an improved training pipeline. The framework includes automated detection head search, structure recombination, model update, as well as model pruning and quantization, achieving efficient and real-time inference on edge devices. This work focuses on the industrial and edge deployment of surface defect detection for chemical fiber yarn rolls. We propose a highly simplified model quantization mechanism, a novel model construction strategy, and an improved training process. The framework includes automatic detection head search, structure reorganization, model update, as well as model pruning and quantization, thereby achieving efficient and real-time inference on edge devices. In addition, we design an integrated posture adjustment device for chemical fiber yarn cakes and build a dedicated dataset for industrial defect detection. To balance open research and laboratory asset management, a representative subset of the dataset is provided, which is sufficient for preliminary training, validation, and further research.In addition, we design an integrated posture adjustment device for chemical fiber yarn cakes and build a dedicated dataset for industrial defect detection. To balance open research and laboratory asset management, a representative subset of the dataset is provided, which is sufficient for preliminary training, validation, and further research. In addition, we have designed an integrated posture adjustment device for the filament yarn bobbins and constructed a dedicated dataset for industrial defect detection. To balance open research and laboratory asset management, a representative subset of the dataset is provided, which is sufficient for initial training, validation, and further research. The released code implements the proposed lightweight network, quantization scheme, and inference pipeline for real-time industrial defect detection. Researchers can use this code and partial dataset for reproduction and further development.The released code implements the proposed lightweight network, quantization scheme, and inference pipeline for real-time industrial defect detection. Researchers can use this code and partial dataset for reproduction and further development. The released code implements the proposed lightweight network, quantization scheme, and inference process for real-time industrial defect detection. Researchers can use this code and part of the dataset for reproduction and further development.