Defect Detection Using Deep Neural Networks and Algorithms: A Survey
Abstract
A defect is an anomaly or an unusual pattern on product surfaces such as on wood surface that must be identified during quality control process. Automatic defect detection is concerned with the problem of finding abnormal patterns on surfaces that usually have uniform patterns or colors. In this paper we review the most recent methods that have been proposed for defect detection in industrial quality control applications. These defect detection methods use general object detection algorithms in combination with deep neural network architectures. Most of these methods uses deep convolutional neural networks are applied to solve different kind of defects detection such as wood, textile, plastic etc. Each method focuses on different types of defects on different kind of surfaces. The paper also reviews the most advanced deep neural networks, how these networks are combined with more sophisticated detection algorithms to approach any defect detection problems in industry. With these new methods it is possible to achieve high accuracy with large scale labeled image datasets. Furthermore, we will also sketch some future research work ideas.