Wood Crack Detection Based on Data-Driven Semantic Segmentation Network
Ye Lin, Zhezhuang Xu, Dan Chen, Zhijie Ai, Yang Qiu, Yazhou Yuan
Abstract
Dear Editor, This letter is concerned with wood crack detection which is important to guarantee the quality of wooden products. In the wood industry, the crack detection is one of the most challenging tasks in the wood defects detection, since the detection accuracy may be reduced due to the stains on the boards, the tiny cracks, and some cracks that are similar to the sound region. To overcome these challenges, we propose a data-driven semantic segmentation network based on U-Net, which is called WCU-Net, for wood crack detection. Specifically, a position attention mechanism is firstly proposed to exaggerate the wood crack positions, then the feature enhancement mechanism is designed to selectively derive more diluted information of tiny crack. Moreover, a residual block is adopted to obtain and fuse multi-scale receptive fields for finding more crack areas that are similar to the sound region. The experimental results show that WCU-Net can improve the accuracy of wood crack detection.