基于深度卷积神经网络的食品包装缺陷检测算法研究

Research on food packaging defect detection algorithm based on deep convolutional neural network

在薯片流水线生产中,如果直接使用YOLOv5网络进行包装缺陷目标检测,精度不够高、训练时间偏长,于是提出了一种基于SENet-YOLOv5的食品包装缺陷目标检测方法。首先,从数据增强这一角度出发,改用了基于Canny边缘检测算法对数据进行处理。然后,在原先DarkNet-53主干网络中融合SENet网络,把重要的特征进行强化来提升准确率。其次,将3处传统卷积层替换为深度可分离卷积层,减少参数量和计算量,最后训练候选区域数据,从而精确地实现定位与分类缺陷。仿真训练结果得出:SENet-YOLOv5模型的检测精度与速度皆得到了提升,对薯片食品包装缺陷的检测准确率为94.6%,检测的平均精度均值(mAP)达到了94.8%,相较干YOLOv5算法提高了7.9个百分点,识别速度大幅度提升。表明所提SENet-YOLOv5缺陷检测方法可应用于薯愿包装检测以提高企业的工作效率。

In the production of potato assembly line,if YOLOv5 network is used to detect packaging defect target directly,the accuracy is not high enough and the training time is too long.Therefore,a food packaging defect target detection method based on SENet-YOLOv5 is proposed.First of all,from the point of view of data enhancement,Canny edge detection algorithm is used to process the data.After that,SENet network is integrated into the original DarkNet-53 backbone network,and important features are strengthened to improve accuracy.Secondly,three traditional convolution layers are replaced by depth separable convolution layers to reduce the number of parameters and computational complexity.Finally,candidate region data is trained to accurately locate and classify the defects.The experimental training results show that the detection accuracy and the speed of SeNet-YOLOv5 model are improved,and the detection accuracy of potato food packaging defects is 94.6%,and the average accuracy(mAP)of the detection reaches 94.8%,compared with the dry YOLOv5 algorithm.That′s a 7.9 percent increase,which is a significant increase in the speed of recognition.The results show that the proposed SeNet-YOLOv5 defect detection method can be applied in potato packaging detection to improve the working efficiency of enterprises.