基于深度学习的工业外观检测通用方法研究
Research on General Method of Industrial Surface Detection Based on Deep Learning
赵梦逸1
作者信息
- 1. 泰州学院信息工程学院,江苏 泰州 225300
- 折叠
摘要
产品外观检测是自动化生产的重要环节,检测准确率和速度是评价自动检测系统效率的主要指标.基于深度学习的外观检测能够满足复杂图像背景下的实时性要求,但通用性有所欠缺.针对此问题,提出了缺陷分割和分类的通用方法.该方法通过语义分割能够精确定位缺陷,并结合残差网络进一步对缺陷进行分类,满足产品缺陷分类要求;也可以仅用分类网络满足仅对产品进行分类的需求.实验结果表明,提出的方法在摄像头模组缺陷检测、药包彩盒缺陷检测以及西红柿分类上满足实时性、通用性和高准确率.
Abstract
Product surface detection is an important part of automatic production.The detection accuracy and speed are the main indexes to evaluate the efficiency of automatic inspection system.Surface detection based on deep learning can meet the real-time requirements under complex image background,but it is lack of universality.To solve this problem,a gen-eral method of defect segmentation and classification was proposed.This method can accurately locate the defects through semantic segmentation,and further classify the defects by combining the residual network to meet the requirements of product defect classification.It is also possible to meet the need of classifying only products with classification networks only.
关键词
机器视觉/深度学习/语义分割/残差网络Key words
machine vision/deep learning/semantic segmentation/ResNet引用本文复制引用
出版年
2024
NETL
NSTL
万方数据