基于卷积神经网络的木材缺陷识别  

Wood defects recognition based on the convolutional neural network

作　　者：徐姗姗[1] 刘应安[1] 徐昇[1] 

机构地区：[1]南京林业大学信息科学与技术学院,江苏南京210037

出　　处：《山东大学学报（工学版）》2013年第2期23-28,共6页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(30671639);江苏省自然科学基金资助项目(BK2009393)

摘　　要：为提高木材缺陷识别率,提出一种基于卷积神经网络算法的识别方法。采用渐近式学习方法来确定训练样本数目,给出了对应的网络结构,降低了算法消耗的时间。试验结果表明,该方法无需对图像进行复杂的预处理,能识别多种木材缺陷,精度较高且复杂度较小,具有很好的鲁棒性,也克服传统算法的诸多固有缺点。To improve the efficiency of wood defects identification,a method based on the convolutional neural network was proposed.A convolutional neural network was presented to recognize the wood defect,and the numbers of training samples were determined by an incremental learning method;the corresponding network structure was designed,and the time consumption could be reduced.Experimental results showed that the pre-processing of a complex image was not needed,and the multi-class defects could be recognized with high accuracy,small complexity and good robustness,while the inherent shortcomings of the traditional algorithm were overcame.

关 键 词：木材缺陷 卷积神经网络 渐进式 图像处理 学习方法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]