文献类型：会议论文

英文题名：An optimized wood knots recognition scheme based on double detection

作者：Wang, Xiao[1]

第一作者：王霄

机构：[1] Research Institute of Wood Industry, Chinese Academy of Forestry, Beijing, China

会议论文集：Proceedings - 2022 International Conference on Computer Engineering and Artificial Intelligence, ICCEAI 2022

会议日期：July 22, 2022 - July 24, 2022

会议地点：Shijiazhuang, China

语种：英文

外文关键词：Color - Decision making - Deep learning - Defects - Network architecture - Object detection - Wood

年份：2022

摘要：To improve the accuracy of wood knots recognition based on computer vision especially the live knot which got similar color with the surroundings, an optimized network architecture was proposed in this study. The model we created was consisted of two parallel networks to cope with two kinds wood defects: light and dark to improve the robustness of the model in color diversity. It was thought that the two well trained networks would be capable of recognizing of any kinds of wood defects between the two typical colors. To avoid repeated recognition of one object cause by the double detection, a decision making mechanism was established based on classification scores to determined the right category. The model was validated by using wood plates which contained 4 kinds of knots. It was shown that the double detection could successfully recognize live knots as well as other defects. The increased accuracy could be attributed to the using of parallel networks which enhanced the implicit feature of live knots. Instead of using complex CNN architecture like some classic models, the proposed method increased the accuracy of wood defects recognition by using parallel networks with relatively simple structure, which increased the speed of model thus was more suitable for industrial application. ? 2022 IEEE.