Comparative Analysis of Object Detection Algorithms for Wood Defect Detection
No TL;DR found
Abstract
This study aims to evaluate the performance of YOLOv8, Faster R-CNN, and Mask R-CNN models in wood defect identification. The dataset, collected in collaboration with a timber company, includes 441 images of wood defects such as knots, oil streaks, black dots, and cracks. These images were annotated using the Computer Vision Annotation Tool (CVAT). The study applied the three models to detect and localize wood defects, comparing their accuracy, detection speed, and precision. Faster R-CNN showed the fastest convergence and the best training results within 24 epochs, while YOLOv8 demonstrated robustness in complex backgrounds but slower mAP improvement. The results indicate the strengths and weaknesses of each model, providing insights for selecting the most suitable model based on specific application needs, and contributing to the development of efficient and automated wood defect detection methods.