?作者：張翠1，楊志清1，周茂杰2（1.桂林理工大學博文管理學院，廣西桂林541006；2.桂林理工大學，廣西桂林541004）
摘要：利用深度學習方法進行塑料產品缺陷檢測時，存在訓練樣本不足，網絡層次增加產生的梯度消失和梯度爆炸等問題，采用殘差網絡ResNet和壓縮與激勵網絡SENet相結合，構建深度學習模型，解決梯度消失、梯度爆炸和注意力分布的問題，利用工業生產中的產品圖像進行缺陷檢測試驗，經過2種試驗結果分析，該算法有效提高了產品缺陷檢測的準確率和召回率。
關鍵詞：ResNet；SENet；缺陷檢測；塑料產品；深度學習
中圖分類號：TG76 文獻標識碼：B 文章編號：1001-2168（2020）11-0013-05
DOI：10.16787/j.cnki.1001-2168.dmi.2020.11.003
Defect detection based on double channel convolutional neural network method
ZHANG Cui1, YANG Zhi-qing1, ZHOU Mao-jie2 (1.Bowen College of Management, Guilin University of Technology, Guilin, Guangxi 541006, China; 2.Guilin University of Technology, Guilin, Guangxi 541004, China)
Abstract: In the process of plastic product defect detection by using deep learning method, there were some problems, such as insufficient training samples, gradient disappearance
and explosion caused by increasing network level, etc.. Combined ResNet with SENet to build a deep learning model, it solved the problems of gradient disappearance and explosion, and attention distribution. Images in industrial production were used to conduct defect detection experiments. Through the analysis of two experimental results, the algorithm could effectively improve the accuracy rate and recall rate of product defect detection.
Key words: ResNet; SENet; defect detection; plastic products; deep learning