基于集群算法优化BP神经网络的NIRS树种识别研究  

NIRS Tree Species Identification Based on Cluster Algorithm Optimized BP Neural Network

作　　者：明曼曼 陈芳[1] 孙恺琦 顾崎岩 吴思齐 王学顺[1] MING Man-man;CHEN fang;SUN Kai-qi;GU Qi-yan;WU Si-qi;WANG Xue-shun(Beijing Forestry University,Beijing 100083,P.R.China)

机构地区：[1]北京林业大学,北京100083

出　　处：《西部林业科学》2020年第5期124-128,共5页Journal of West China Forestry Science

基　　金：北京林业大学大学生创新创业训练项目(G201910022066);中央高校基本科研业务费专项基金(2015ZCQ-LY-01)。

摘　　要：为探究基于近红外光谱分析技术的木材树种准确、快速识别新方法,并研究光谱波段范围对识别的影响,以大叶桉、杉木、落叶松、马尾松、樟子松5种木材样品为研究对象,针对3种光谱波段范围,分别建立未优化的BP神经网络模型(BP-ANN)、粒子群算法优化BP神经网络模型(PSO-BP)和人工蜂群算法优化BP神经网络模型(ABC-BP),对比模型识别准确率和运行时间。研究结果显示,波段越长,模型的识别准确率越高;PSO-BP与ABC-BP模型的识别准确率均高于BP-ANN,其中ABC-BP模型的识别效果最好,可达到95.333%;ABC算法较于PSO算法优化BP神经网络模型用于木材近红外光谱树种识别时间更短。基于集群算法优化BP神经网络模型能有效应用于树种识别研究,具有一定应用前景。In order to explore a new method for accurate and rapid identification of wood species based on near-infrared spectroscopy analysis technology,and study the impact of spectral band range on identification,an unoptimized BP neural network model(BP-ANN)and a particle swarm optimization algorithm were established for the three spectral band ranges to optimize the BP neural network.Five kinds of wood samples,namely eucalyptus,fir,larch,masson pine,and scotch pine were selected as research objects.The network model(PSO-BP)and artificial bee colony algorithm optimize the BP neural network model(ABC-BP),and compare the model recognition accuracy and running time.The research results showed that the longer the band,the higher the recognition accuracy of the model.The recognition accuracy of the PSO-BP and ABC-BP models is higher than that of BP-ANN,of which the ABC-BP model has the best recognition effect,which can reach 95.333%.Compared with the PSO algorithm,the ABC algorithm in the optimized BP neural network model can be used to identify tree species in the wood near-infrared spectrum in a shorter time.In general,optimized BP neural network model based on the cluster algorithm can be effectively applied to the research of tree species identification and has certain application prospects.

关 键 词：近红外光谱技术 木材树种识别 BP神经网络 人工蜂群算法 粒子群算法 

分 类 号：S781[农业科学—木材科学与技术] O657.3[农业科学—林学]