面向智能手机玻璃盖板缺陷检测的YOLOv3改进和应用

Improvement and application of YOLOv3 for defect detection of smart phone glass covers

针对智能手机玻璃盖板缺陷检测方法存在检测柔性差、良率低、检测时间长等问题,提出一种改进YOLOv3的智能手机玻璃盖板缺陷检测方法。在特征提取网络方面增加通道注意力机制以解决缺陷特征不明显的问题,在特征检测网络方面增加了104×104维度大小的特征图以解决缺陷多尺度的问题,最后对模型进行剪枝减少模型参数,提高缺陷检测速度。从智能手机玻璃盖板生产现场获得涵盖崩边、坑点、脏污和划痕等4类缺陷的图片构建缺陷数据集,对本文提出的方法和Faster R-CNN、YOLOv3、YOLOv4等算法进行对比实验和分析。实验结果表明,本文提出方法的检测平均精度均值(mean Average Precision,mAP)为81.0%,检测速度为43.1 fps。相比原始YOLOv3算法,检测mAP提升了3%,检测速度增加了6.7 fps,相比于其他深度学习算法,检测速度和检测精度均有所提升。所提方法满足智能手机玻璃盖板工业生产现场缺陷高精度、高效检测的需要。

To address the problems of poor detection flexibility,low yield rate and long detection time of smartphone glass cover defect detection methods,an improved YOLOv3 defect detection method for smartphone glass cover is proposed.A channel attention mechanism is added to the feature extraction network to solve the problem of inconspicuous defect features,a feature map of 104×104 dimensional size is added to the feature detection network to solve the problem of multi-scale defects,and finally the model is pruned to reduce the model parameters to improve the defect detection speed.The defect dataset is constructed by obtaining images covering four types of defects,such as chipped edge,pit,dirty and scratches,from the production site of smartphone glass cover.The proposed method and algorithms such as Faster R-CNN,YOLOv3 and YOLOv4 are compared for experiments and analysis.The experimental results show that the detection mAP(mean average precision)of the proposed method is 81.0%and the detection speed is 43.1 fps.Compared with the original YOLOv3 algorithm,the detection mAP is improved by 3%and the detection speed is increased by 6.7 fps.Compared with other deep learning algorithms,the detection speed and detection precision are improved.The proposed method meets the need for high-precision and efficient detection of defects in the industrial production site of smartphone glass covers.