基于点集匹配的缺陷样本图像生成方法  

Method of defect sample image generation based on point set matching

作　　者：高海洋 张明川 葛泉波 刘华平[3] GAO Haiyang;ZHANG Mingchuan;GE Quanbo;LIU Huaping(School of Information Engineering,Henan University of Science and Technology,Luoyang 471023,China;School of Automation,Nanjing University of Information Science and Technology,Nanjing 210044,China;Department of Computer Science and Technology,Tsinghua University,Beijing 100084,China)

机构地区：[1]河南科技大学信息工程学院,河南洛阳471023 [2]南京信息工程大学自动化学院,江苏南京210044 [3]清华大学计算机科学与技术系,北京100084

出　　处：《智能系统学报》2023年第5期1030-1038,共9页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金重点项目(62033010);中原科技创新领军人才项目(224200510004)。

摘　　要：针对工业缺陷检测中存在的由产品次品率过低、产品迭代更新过快、缺陷种类难以覆盖全部以及缺陷样本高质量标注难度较高导致的小样本问题,使用基于点集匹配的缺陷样本图像生成方法来对缺陷数据进行扩充。将缺陷部位从多特征角度进行变换,使用单张样本进行扩充得到不同特征的缺陷图像,解决小样本条件下深度学习方法难以生成高质量缺陷图像的问题。通过图像评估与实验验证,该方法生成的图像具有更好的视觉效果,并且对缺陷与分割模型有着高效的提升。该方法可应用于样本较少的深度学习模型训练过程中,达到扩充样本提高训练效果的目的。The paper presents a novel approach for generating defect sample images using point set matching,which ad-dresses the challenges posed by small-sample in industrial defect detection.These challenges arise due to low defective rates of products,rapid iterative updating of products,limited coverage of defect types,and difficulty in obtaining high-quality labeled defect samples.The proposed method transforms defects from a multifeature perspective and applies a single-sample expansion technique to generate defect images with diverse characteristics.This method solves the prob-lem of the difficult generation of a high-quality defect image by deep learning under small-sample conditions.Through image evaluation and experimental verification,this method can produce images with superior visual effects and can ef-fectively improve defect segmentation and detection.This method can be applied to the training process of the deep learning model with few samples for sample expansion and improvement of the training effect.

关 键 词：工业 缺陷检测 小样本问题 点集匹配 样本扩充 缺陷样本生成 有效训练 循环生成对抗网络模型 矢量化变分自动编码器 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]