Abstract: Several object detection problems inevitably occur when the product components are detected in the packaging procedure of the traditional industrial production line, including slow detection speed, low-level automation, and low detection accuracy. To solve these problems, a vision-based industrial objects detection system is established to automatically inspect the product components, which is based on deep learning. Firstly, an experimental platform is designed to obtain the images containing product components. The convolution layer structures shared by the region proposal network and region convolutional neural network are modified. Thus, a novel object detection method is proposed to accurately locate the product components, which can adaptively detect different product components with different shapes and different sizes by means of an end-to-end mode. The experimental results show that the established system is superior to the existing detection methods, which has the advantages of high detection speed and high detection accuracy.

Key words: industrial objects, deep learning, Region Proposal Network and Region Convolutional Neural Network（RPN+RCNN）, location detection

摘要： 针对零部件在工业生产线包装过程中存在检测速度慢，自动化检测水平低下，检测准确率不高等问题，提出一种基于深度学习的工业零部件检测系统的方案设计，实现对零部件自动检测的功能。设计一种实验检测平台用于获取包含待检测零部件的图像；提出一种改进网络共享卷积层结构的方法，融合区域建议网络和区域卷积网络建立一种检测方法对目标零部件进行准确定位识别，适应各种形状大小不一的零部件检测，实现端到端训练输出定位检测结果。实验结果表明，系统的检测效果与传统的检测方法相比，具有速度快，检测准确率高等优越性。

关键词: 工业零部件, 深度学习, RPN+RCNN检测网络, 定位检测