摘要
针对木材分类特征的繁多,同科属木材种间差异小,单纯依靠一种特征分类易误识等特点,提出了一种基于模糊BP综合神经网络的新型木材树种分类识别方法.该方法使用分类特征的模糊化处理,充分考虑木材的分类特征本身存在的不确定性;同时使用一种特征级数据融合方法,该综合神经网络包括颜色特征、主要和次要纹理特征和光谱特征4个BP子网络;并用散度进行光谱特征波段的特征选择,还运用遗传算法对网络结构进行优化处理,提高了该综合神经网络的收敛性和稳定性.实验时针对东北地区常见的5种树种(白松、樟子松、落叶松、杨木和桦木)木材进行分类测试,实验结果表明,5种树种木材的混合识别率达到89%,具有较好的分类识别精度.
Wood species identification with only one feature lacks accuracy since the difference is small for many features of the same wood species. A novel wood recognition scheme was proposed based on fuzzy BP overall neural network,which had the following advantages. First,the classification features were blurred to deal with the uncer-tainty of wood features. Second,a feature-level data fusion scheme was used so that the neural network consisted of 4 BP sub-networks concerning color feature,texture feature and spectral feature. Finally,a feature selection procedure on the spectral feature interval was applied by use of divergence and the genetic algorithm was used to optimize the network structure so as to improve the network’s convergence and stability. Experiments on the 5 ordinary wood spe-cies in northeast region of China indicated that the overall recognition rate reached as high as 89% for 5 wood species,showing good recognition accuracy.
出处
《天津大学学报(自然科学与工程技术版)》
EI
CAS
CSCD
北大核心
2015年第2期147-154,共8页
Journal of Tianjin University:Science and Technology
基金
教育部新世纪优秀人才支持计划专项资助项目(NCET-12-0809)
教育部中央高校基本科研业务费专项基金资助项目(2572014EB05-01)
中国博士后科学基金特别资助项目(2012T50318)
关键词
模式识别
树种识别
特征选择
数据融合
光谱分析
pattern recognition
wood species recognition
feature selection
data fusion
spectral analysis