摘要
应用图像识别技术实现工件疵病自动检测可以提高效率,降低人工成本。卷积神经网络(CNN)具有很强的特征提取能力,广泛应用于图像识别等领域。但是,已提出的网络模型普遍存在参数量和计算量巨大,以及海量工件数据集中、人工标注成本高等缺点,很难应用于工件疵病的实时自动识别。本文提出了一种基于轻量级CNN与主动学习的工件疵病识别方法,该方法通过深度可分离卷积和反转残差卷积构建一种轻量级卷积神经网络,并在识别过程中采用主动学习方法不断添加标注样本。实验表明,提出的方法识别精度达到98.3%,并且能节省18.8%的人力标注成本。
Automatic detection of workpiece defects through image recognition technology can improve efficiency and reduce labor cost.Because of its strong feature extraction ability,convolutional neural networks(CNN)are widely used in image recognition and other fields.However,the existing CNN models are difficult to be applied to automatic detection of workpiece defects in real time with the disadvantage of huge amount of parameters and computation,and the high cost of manual labeling in massive workpiece data sets.In this paper,a new approach for workpiece defect detection is proposed by combining lightweight CNN with active learning,which constructs a lightweight CNN by deep separable convolution and inverted residual convolution,and uses active learning method to add labeled samples continuously in the process of detection.Experiments show that the proposed method can achieve a high recognition accuracy of 98.3%and a labeling cost reduction of 18.8%simultaneously.
作者
姚明海
杨圳
Yao Minghai;Yang Zhen(College of Information Engineering, Zhejiang University of Technology, Hangzhou 310023)
出处
《高技术通讯》
EI
CAS
北大核心
2020年第4期325-332,共8页
Chinese High Technology Letters
基金
国家自然科学基金(61871350)资助项目。
关键词
卷积神经网络(CNN)
主动学习
轻量级
疵病识别
convolutional neural network(CNN)
active learning
lightweight
defect recognition
