基于支持向量机的木材树种识别模型  

Identification of wood species using support vector machine

作　　者：骆立 徐兆军 王晓羽 周康 那斌 LUO Li;XU Zhaojun;WANG Xiaoyu;ZHOU Kang;NA Bin(College of Materials Science and Engineering,Nanjing Forestry University,Nanjing 210037,China)

机构地区：[1]南京林业大学材料科学与工程学院,南京210037

出　　处：《林业工程学报》2022年第4期122-127,共6页Journal of Forestry Engineering

基　　金：国家重点研发计划(2016YFD0600703)。

摘　　要：目前林业信息化正由数字林业迈向智慧林业,高效、无损的木材树种识别技术有利于推动我国林业信息化、智能化发展的进程。为了满足市场对木材高效精准识别的需求,将木材树种识别问题转化为多分类问题,开发了一种基于支持向量机结合线性降维算法的木材树种识别模型。具体而言,首先采用无监督的主成分分析和有监督的线性判别分析,分别对木材近红外高维光谱数据进行降维处理;其次将降维后的特征输入支持向量机模型中,输出各个树种类别上的概率分布。借助网格搜索法并结合5折交叉验证法选取最优核函数和核函数参数,探讨了支持向量机不同核函数对于木材树种分类效果的影响。为了评价模型的识别能力,选取准确率、混淆矩阵和ROC曲线评价提出的模型,并进一步探讨了本木材树种识别方法的可行性。实验结果表明,利用近红外光谱特征的支持向量机模型能准确且高效地识别木材树种。其中,线性判别分析结合支持向量机的模型分类准确率可达97.54%,模型运行速率为6.53 s。At this stage, forestry informatization is moving from digital forestry to intelligent forestry in China. Wood identification plays a key role in wood utilization not only for determining the appropriate use but also for supporting the legal timber trade. With the rapid development of the economy in China, the demands for the quantity and quality of wood have increased year by year, and therefore, exploration of an effective testing method for identification of wood species is of great significance to help us make better use of wood. For this reason, a novel method using the near-infrared spectroscopy technique associated with support vector machine(SVM) was proposed for the discrimination of five wood species(Pterocarpus soyauxii, Pterocarpus tinctorius var. chrysothris, Hevea brasiliensis, Quercus mongolica and Fraxinus mandshurica). Specifically, the spectra were obtained in the region from 900 to 1 650 nm using 1 000 samples from five different species firstly. The unsupervised principal component analysis(PCA) and supervised linear discriminant analysis(LDA) were performed on the spectral signatures of wood. Secondly, wood species identification models were built from the de-noised spectra using PCA-SVM and LDA-SVM methods, respectively. In addition, the grid search combined with five-cross-validation was applied to analyze the best combine parameters of models. Finally, the accuracy, confusion matrix, and ROC curve were selected to evaluate the proposed model. Results indicated that LDA performed better compared with PCA. This could be due to PCA did not separate them according to their types. This study aimed at comparing the rate of correct predictions of species with PCA-SVM and the LDA-SVM models and evaluate the potential of the two classifiers for the discrimination of wood species. In conclusion, the two models were adequate to classify and identify the five wood species based on the respective NIR spectra with simple, rapid, and non-destructive advantages. Particularly, it should be emphasized that

关 键 词：支持向量机 近红外光谱 树种识别 主成分分析 线性判别分析 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]