基于深度学习的织物疵点检测研究进展

Research progress in fabric defect detection based on deep learning

为提高疵点检测的准确性和通用性,实现使用简洁而有效的形式对织物图像的特点和疵点的本质特征进行综合表达,首先,介绍了深度学习技术,对引入了深度学习的疵点检测方法进行综述,同时对深度学习与疵点检测的内在关系进行阐述;然后,分析总结了深度学习的概念及代表性的计算模型,并对引入深度学习的疵点检测方法进行归纳、总结和分类;最后,对典型的方法进行了分析,讨论了各种方法的优缺点,并对未来的研究趋势进行了展望。指出:随着深度学习的发展,探索更加通用的检测方法是推进深度学习在织物疵点检测领域应用的努力方向。

Significance With the development of science and technology,the improvement of product quality is highly demanded.Although the technologies used in producing textile products have undergone revolutionary advancement which contributes to the textile quality dramatically,defects in textile products such as fabrics remain to be a reality.Fabric defect detection plays an important role in textile industry,and fabric defect detection technology based on deep learning has been paid more and more attention.This paper reports a research and development progress in fabric defect detection based on deep learning.Progress Deep learning is mainly composed of four steps,i.e.,defining model and loss function,training the model,finding optimization method and loop iteration.The research focus for fabric defect detection method based on deep learning mainly includes deep learning models such as convolution neural network(CNN)and automatic encoder(AE).The stack denoising automatic encoder based on Fisher criterion introduces deep learning into this field for the first time,which provides a new idea for the application of deep learning to the field of defect detection.Convolution neural network has achieved good results in the field of image recognition because of its strong nonlinear fitting ability.More precision-based detection algorithms based on candidate regions and more speed-based algorithms based on regression analysis are present.While the advantages of convolution neural network is exploited,other methods are used for exploring the possibility of combined use of models,and provide new ways for defect detection.Conclusion and Prospect Fabric defect detection methods based on deep learning in recent years are reviewed and summarized,and the effects of different models are compared in detail.Advantages,disadvantages and applicable scope of each model are analyzed,and future development of fabric defect detection method based on deep learning model is prospected.Deep learning models can improve the detection efficiency,but still have some deficiencies.In order to optimize the accuracy of fabric image defect detection,breakthrough should be made from the following aspects in the future.1)High quality data sets should be established.2)Specific evaluation criteria need to be established.3)The applicability should be extended.A single detection method often has limitations,but when different defect detection methods are utilized to deal with different detection needs,the detection results are often different,therefore hybrid methods would have better applicability.