基于改进YOLOv3的锌阴极板残留物图像识别方法

为识别剥锌机锌阴极板的残留物,研究了一种基于深度学习的阴极板残留物图像识别方法。在YOLOv3单阶段图像检测模型的基础上,将原始网络结构Darknet53替换成具有深度可分离卷积的Xception网络结构来提升检测模型的特征提取能力,通过改进候选框生成策略来减轻训练过程中正负样本不平衡。实验从训练过程损失曲线以及识别准确率分别对比了改进结构与原始结构的识别效果。结果表明,基于改进YOLOv3的方法提升了检测模型的泛化能力,对阴极板残留物的识别准确度高达95%,较原始的YOLOv3模型提升了3%。

Image recognition method of zinc cathode plate residue based on improved yolov3

In order to identify the residue of zinc cathode plate of zinc stripping machine, this paper studies an image recognition method based on deep learning. Based on the yolov3 single-stage image detection model, the original network structure darknet53 is replaced by the Xception network structure with deep separable convolution to improve the feature extraction ability of the detection model, and by improving the candidate box generation strategy to reduce the imbalance between positive and negative samples. The experiment compares the recognition effect of the improved structure and the original structure from the loss curve of training process and the recognition accuracy. The results show that the method based on improved yolov3 improves the generalization ability of the detection model, the recognition accuracy of cathode plate residue is up to 95%, 3% higher than the original yolov3 model.