Object Detection using Deep Learning in a Manufacturing Plant to Improve Manual Inspection
Afshin Rahimi, Mohammad Anvaripour, Khizer Hayat
Abstract
Industrial inspection to ensure high quality of products is one of the challenging tasks in manufacturing. Since humans cannot provide a fully reliable inspection compared to a machine, an intelligent system can complement the human inspection process in many industrial applications. Therefore, this paper provides a real industrial case-study for applications of deep learning and computer vision in increasing the quality of inspection by providing a monitoring system for managers to detect inspection flaws in the automotive industry. To achieve this, an enhanced deep neural network approach is proposed to detect different classes of objects in each frame of an inspection video stream. The sequence of the detected classes is then compared to the manufacturer's inspection process to identify potential flaws in inspection process compliance. A comprehensive comparison of the proposed approach's performance to other object detection methods for training speed and detection accuracy proves its superiority Compared to the other state-of-the-art methods.