一种基于深度学习的产品分类统计方法的研究  

Research on Statistical Method of Product Classification Based on Deep Learning

作　　者：王占云[1] 闫志华[1] WANGZhan-yun;YANZhi-hua(School of Mechanical Engineering,Zhengzhou University,He’nan Zhengzhou450001,China)

机构地区：[1]郑州大学机械工程学院,河南郑州450001

出　　处：《机械设计与制造》2020年第3期163-166,共4页Machinery Design & Manufacture

摘　　要：对于流水线上多类产品不规则混合放置条件下的统计问题,利用基于faster R-CNN和改进的深度卷积网络结合的图像识别方法,实现了对产品的在线分类计数。首先利用faster R-CNN方法获取图像中产品的位置信息,然后用改进的深度卷积网络对图像进行特征提取,将产品区域的位置信息映射到最后一层特征图上,再用ROI池化对产品区域特征进行尺度归一化,把归一化后的产品特征输入softmax分类器识别,从而实现了分类统计。实验结果表明,该方法能够在工业生产中实现多类混线产品在线自动分类统计。For the statistical problems of irregular products mixed with multiple products on the production line,we use the image recognition method based on the faster R-CNN and the improved deep convolution network to achieve the classification and counting of products in the production line. Firstly,we use the faster R-CNN method to obtain the location information of the product in the image,and then we use the improved convolutional neural networks to extract image feature. Then the location information of the product area is mapped to the last layer of feature map,the scale normalization of the product area features is carried out by ROI pool. The normalized product features are entered into softmax classifier recognition,and the classification statistics of the products are finally realized. The experimental results show that the method can realize online classification and statistics of multi class collinear products in industrial production.

关 键 词：深度学习 分类统计 图像识别 FASTER R-CNN ROI池化 

分 类 号：TH16[机械工程—机械制造及自动化] TP29[自动化与计算机技术—检测技术与自动化装置]