An Identification Method for Mechanical Parts Using on Deep Learning with Single Shot-multibox Detector MobileNet
Konferenz: MEMAT 2022 - 2nd International Conference on Mechanical Engineering, Intelligent Manufacturing and Automation Technology
07.01.2022 - 09.01.2022 in Guilin, China
Tagungsband: MEMAT 2022
Seiten: 5Sprache: EnglischTyp: PDF
Autoren:
Liu, Junfeng; Fan, Yong; Liang, Zhenghao; Fu, Liye (School of Mechanical and Electrical Engineering, Guilin University of Electronic Technology, Guilin, Guangxi, China)
Wu, Shuqin (School of Architecture and Transportation Engineering, Guilin University of Electronic Technology, Guilin, Guangxi, China)
Xie, Wu (School of Computer Science and Information Security，Guilin University of Electronic Technology, Guilin, Guangxi, China)
Inhalt:
SSD (Single Shot-multibox Detector)-MobileNet is widely used in embedded systems and intelligent manufacturing systems due to its lightweight structure and acceptable precision loss. In order to identify mechanical parts, images of four parts are collected, labelled and identified based on SSD-MobileNet with deep learning model. The TensorFlow frame-works are used only for training, and the trained models are applied to implement real-time detection. The results show that the model has a good recognition effect on mechanical parts with the accuracy of 0.936 and the recall rate of 0.956 under the conditions of not large-scale quantity of parts, which is useful to meet the identification and detection of parts to achieve the application purpose of parts classifications.