An Optimized Lightweight Model for Real-Time Wood Defects Detection based on YOLOv4-Tiny

Abstract

Many wood manufacturers are still relying on manual human eyes inspection for wood defects detection. This approach is tedious, inconsistent, inefficient, and prone to human errors. Machine vision technology can provide a satisfactory solution for wood defects detection and reduce the manpower required. In this paper, a lightweight object detection model is proposed for the detection of four types of wood defects based on the YOLOv4-Tiny architecture. The accuracy of the model is improved by modifying the loss function for YOLOv4-Tiny to incorporate Intersection over Union into its objectness loss. The results showed that the improvement made has successfully enhanced the model’s accuracy and the best model can achieve a mean average precision of 88.32% running at 225.22 frames per second.