Surface Flaw Detection of Industrial Products Based on Convolutional Neural Network

2018 4th International Conference on Environmental Science and Material Application

生态环境科学;材料科学

Zhang, Yongjun^1 ; Wang, Ziliang^1

Institute of Information Photonics and Optical Communications, BUPT, Beijing

100876, China^1

关键词: Batch sizes;    Classification accuracy;    Convergence rates;    Convolution kernel;    Convolutional neural network;    Industrial product;    Surface flaw;    Typical application;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/252/2/022114/pdf
DOI  :  10.1088/1755-1315/252/2/022114

来源: IOP

【 摘 要 】

Surface flaw detection in industrial products is a typical application of image classification. By improving the structure of Convolutional Neural Network (CNN), for example, the first large-scale convolution kernel is replaced by a cascaded 3×3 convolution kernel; replaces the whole with a 1×1 convolution kernel and Global Average Pooling Connection layer; sets the appropriate batch-size, the convergence rate and convergence accuracy of the model are greatly improved. Experiments show that the proposed method has a classification accuracy of more than 96% in the detection of automotive hose surface flaws.