Hand-detection with Transferrable Design for Smart Factories

Abstract

Nowadays, a smart factory in Industry 4.0 often must produce a variety of products, so its assemblers need to learn different assembly processes and post-inspections. Smart cameras that leverage edge computing (hereinafter referred to as edge cameras) can now incorporate deep neural networks (DNNs) and have been widely used in smart factories. However, in response to the demand for rapid learning and deployment of DNNs across different assembly lines, which has not been addressed in previous studies, we propose "Knowledge Transfer across Multiple Assembly Lines" (KTaMAL) to transfer learned knowledge across different assembly lines. The experimental results show that the model prediction accuracy of KTaMAL is improved by 8% compared with non-transfer-learning based approaches and the training time can be significantly reduced by approximately 80%.