The Prospect of Artificial Intelligence-Based Wood Surface Inspection: A Review

Abstract

The demand for solid wood is high in the construction and manufacturing industries, and the quality of the wood is crucial. Defects in solid wood can result in hazardous accidents or financial loss. While manual visual inspection of defects is time consuming and labor intensive, Automated Optical Inspection (AOI) systems provide a solution that is hindered by defect variations and environmental factors such as moisture content and lighting conditions. AOI systems coupled with machine learning algorithms have emerged as a promising approach for inspecting wood defects. Despite their promising results compared to manual visual inspection and AOI systems, machine learning algorithms have shown several limitations in terms of complex image processing methods, feature engineering, and hyperparameter dependence. Deep learning algorithms have tremendous potential and have become trends in wood defect inspection in recent years, particularly Convolutional Neural Networks (CNNs), single-shot detectors (SSD), You Only Look Once (YOLO), and faster region-based neural networks (Faster R-CNN) algorithms. The coupling of machine vision technology with deep learning algorithms can enhance the efficiency and accuracy of wood defect inspection, and their impact has been proven in several studies. This study aims to provide a comprehensive overview of wood defect inspection approaches by analyzing related studies on machine learning-based and deep learning-based defect inspection methods. Their principles, procedures, performance, and limitations were compared and discussed. Subsequently, future trends and challenges in wood defect inspection are also discussed to provide a detailed understanding and direction for related fields.

Keywords

• Deep learning
• defect inspection
• machine learning
• machine vision
• wood