小样本学习在纺织品缺陷检测中的新发展

New Developments of Few-Shot Learning in Textile Defect Detection

由于基于深度学习的检测方法是由数据驱动的,因此实现良好的检测效果关键在于拥有充足的训练数据。然而,获取织物缺陷图像通常受到数据不足和样本不平衡等问题的困扰,这直接影响了检测算法的最终效果。因此,需要利用小样本学习技术来应对这一挑战。简要介绍织物疵点图像的主要特点,以便更好地理解小样本学习应用于织物疵点检测领域所面临的挑战。从数据增强、度量学习、元学习、微调4个关键方面探讨小样本学习技术,分析相关技术在工业缺陷检测领域的应用现状,进一步指出小样本学习技术在织物缺陷检测领域中所面临的机遇和挑战,为小样本学习在织物缺陷检测领域的研究提供有价值的参考。

Due to the data-driven nature of deep learning based on detection methods,the key to achieving good detection results lies in having sufficient training data.However,obtaining images of fabric defects is often plagued by issues such as insufficient data and imbalanced samples which directly affect the final effectiveness of detection algorithms.Therefore,it is necessary to utilize fewshot learning techniques to address this challenge.The main characteristics of fabric defect images were briefly introduced to better understand the challenges faced by applying few-shot learning in the field of fabric defect detection.The few-shot learning technique was explored from 4 key aspects:data augmentation,metric learning,meta learning and fine-tuning.The current application status of related technologies in the field of industrial defect detection was reviewed and analyzed.The future opportunities and challenges faced by few-shot learning technique in the field of fabric defect detection were further pointed out,aiming to provide valuable reference for the research of few-shot learning in the field of fabric defect detection.