迭代阀值分割法的木材分类  

Classification of Wood Based on Iterative Threshold Segmentation

作　　者：官飞[1] GUAN Fei(Department of Automation Engineering,Fujian Forestry Vocational and Technical College,Nanping,Fujian 353000)

机构地区：[1]福建林业职业技术学院自动化工程系,福建南平353000

出　　处：《武夷学院学报》2021年第6期35-39,共5页Journal of Wuyi University

基　　金：福建省教育厅A类科技项目(JAS170962)。

摘　　要：运用迭代阀值分割法的手段,将珍贵树种木片图像中的特征区域和图像背景分离,再运用形态学处理方法中的开运算完成木片图像的孔洞、斑点处理,并对图像进行去噪声、平滑处理,得到木材特征图像中的二值图形。把提取到的RGB特征、HSV特征和Gabor特征做为输入量放入设计好的BP神经网络进行训练测试,经过正向传播与误差反向传播多次交替,通过阈值以及权值的不断调整,精确BP神经网络的运算值,运算之后运用混淆矩阵来分析分类效果,通过观察矩阵中对应象限中的TP和TN数量,得出木片特征样本预测分类的识别率情况,从而达到珍贵木材识别的效果。This paper uses the method of iterative threshold segmentation to separate the feature regions and image background of the wood slice images of precious tree species,then uses the open operation in the morphological processing method to complete the hole and spot processing of the wood slice image,and then de-noise and smooth the image to obtain the binary figure in the wood feature image.The extracted RGB feature,HSV feature and Gabor feature are put into the designed BP neural network to be trained and tested.After the positive propagation and the error back propagation,the BP neural network is trained and tested.By observing the number of TP and TN in the corresponding quadrants of the matrix,the recognition rate of the prediction classification of the chip feature samples is obtained,to achieve the effect of precious wood identification.

关 键 词：木材识别 迭代阀值分割 形态学 特征提取 BP神经网络 

分 类 号：TN913[电子电信—通信与信息系统]