Product Defect Detection Using Deep Learning

Abstract

To maximize production efficiency, product quality control is paying more attention to the quick and reliable automated quality visual inspection. Product defect detection is a critical part of the inspection process. Manual defect detection has a lot of flaws that can be overcome using a deep learning approach. In this paper we have proposed and implemented the deep learning models to detect defects in the manufactured product. Two types of classification, i.e., binary and multiclass classification, is done using CNN, AlexNet, and YOLO algorithms. For the binary classification which is just used to check whether there is a defect in the product, we have proposed three different architectures of CNN, out of which the third CNN model gave 99.44% and 97.49% for training and testing, respectively. We also tested the AlexNet model and got accuracy of 97.6%. And for the multiclass classification that is used for identification of type(s) of defects, the YOLOv8 model is proposed and implemented, which gives better results by attaining a remarkable accuracy of 98.7% for multiclass classification. We also designed and developed the Android Application, which is used on the field for defect detection in the manufacturing industry.