Application of mobile injection molding pump defect detection system based on deep learning

• Published in: 2022 2nd International Conference on Consumer Electronics and Computer Engineering (ICCECE) pp. 470 - 475
• Main Authors: Li, Jun-xian, Li, Ying, Zhang, Xue-xi, Zhan, Rui-dian, Tang, Hao-ran, Huang, Hong-min
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.01.2022
• Subjects: Chemical products; Computational modeling; Convolution; Deep learning; deformable convolution; Deformable models; Intelligent injection pumps; NCNN; Quality control; Quantization (signal)
• DOI: 10.1109/ICCECE54139.2022.9712789

As a core product in daily chemical products, injection pumps are used frequently in life, so the quality control of injection pumps has become a very important aspect for injection pump manufacturers. The purpose of this paper is to study the deep learning-based defect detection method for injection pumps to achieve a fully automated, stable, efficient and easy-to-deploy defect detection system for injection pumps. A CenterNet-based defect detection model is proposed to improve the DLA based on deformable convolution to increase the recognition ability of irregular defects, and a focused loss function is used to solve the problem of uneven positive and negative samples. And the model compression method of channel sparsification and the model quantization method of NCNN are used to optimize the deployment of the model in mobile.