Image detection of wood surface defects based on deep learning

Aiming at the defect image detection of live knots, dead knots, wormhole and so on, a wood defect image detection method based on depth learning was proposed. Firstly, by training Faster-RC-NN network, a detection model for locating and recognizing wood defects is obtained. Secondly, the image is denoised by NL-Means method, and image enhancement is achieved by linear filtering, adjusting contrast and brightness. Thirdly, the image is processed by binarization, and the edge feature points of defects are extracted according to the difference of pixel values to realize wood defects fine segmentation. Finally, the ellipse fitting method is improved to realize the ellipse fitting of wood defect edge point set, and a new wood defect processing scheme is provided. The experimental results show that the algorithm has better wood defect location and classification ability, and gets better segmentation and fitting effect. The filling volume of wood can be reduced by about 10 % in the process of defect repair.