基于蜂群单阈值分割的SRC板材缺陷分类方法

Classification method for SRC wooden board defects based on single threshold segmentation of artificial bee colony

针对传统单阈值板材缺陷分割算法易陷入局部最优、早熟以及收敛速度慢等缺点,提出了一种基于改进蜂群算法的单阈值分割算法.为了提高缺陷分类准确率并减少运算量,将稀疏表达分类器(SRC)运用到板材缺陷分类过程中.改进算法每次迭代都会同时进行全局和局部搜索,且侦查蜂随机全局选取蜜源以加快收敛速度,搜索半径可以根据时变搜索参数进行自适应调整,SRC可将缺陷分类问题转换为求最稀疏系数解的过程.结果表明,本文算法可以准确快速地计算出最佳分割阈值,并将分类准确率提高到90%以上,具有一定的可靠性与可行性.

Aiming at such shortages as easily falling into local optimum situation, precocity and slow convergence speed of traditional single threshold segmentation algorithm for wooden board defects, a single threshold segmentation algorithm based on improved artificial bee colony （ABC） algorithm was proposed. In order to improve the defect classification accuracy and reduce the computational work, the sparse representation-based classifier （SRC） was applied in the classification process of wooden board defects. The improved algorithm simultaneously could realize both global and local search during each iteration, and the bee scouts could select nectar resources randomly in global area to speed up the convergence rate. The search radius was adaptively adjusted according to time-varied search parameters, and the SRC transformed the defect classification problem into the problem of obtaining the most sparse coefficient solution. The results show that the proposed algorithm can compute the optimal segmentation threshold, improve the classification accuracy to above 90%, and has certain reliability and feasibility.