Real-Time Plastic Surface Defect Detection Using Deep Learning
Muhammad Ikhsan Roslan, Zaidah Ibrahim, Zalilah Abd Aziz
Abstract
Quality control is a process utilized in the plastic packaging industry to ensure that the products that are produced are high-quality. This is achieved by identifying and eliminating defects before they are commercialized in the market. The quality of plastic surfaces makes a difference in how customers see the final product. To avoid experiencing errors and minimize product defects, manual surface defect detection is typically performed by humans through the naked eyes. Due to slow detection speed, high labor costs, and visual acuity limitations, manual defect detection can no longer meet today's demands. Therefore, real-time identification of plastic surface defects using computer vision technology is required. This paper proposes a method for the real-time detection and classification of plastic surface defects using deep learning which is You Only Look Once (YOLO). YOLO has shown excellent performance in object detection and this research applies YOLOv5. It is performed by training a custom dataset obtained from the plastic packaging industries to identify defective surfaces and at the same time to obtain its detection accuracy in terms of precision, recall, F-measure, and mAP.