基于BP神经网络的森林树种分类研究

Study on forest species classification based on BP neural network

精准的森林树种分类是林业遥感研究的重要课题。高光谱数据具有丰富的波谱信息,能够探测到不同植被光谱的细微差异,为森林树种分类研究提供了数据来源。人工神经网络提供了任意维数输入输出矢量之间的非线性映射,能够逼近任意的非线性连续系统,为森林树种分类研究提供了技术手段。以马尾松成熟林、樟树幼龄林及荷花玉兰幼龄林为分类对象,将高光谱特征参数作为输入矢量,森林树种类别作为输出矢量,构建BP神经网络,开展分类研究。结果表明:马尾松成熟林及樟树幼龄林的分类精度达100.0%,样本分类综合精度可达93.3%。

Precision forest species classification is a significant project in forestry remote sensing.Hyperspectral datas contain rich spectral information and can detect small differences of spectrum from different types of vegetation.Artificial neural network provides technical means for study on forest species,because it can provide nonlinear mapping of the input eigenvector and output eigenvector with arbitrary dimension,and also can approximate to any nonlinear continuous system.The BP neural network based on mature Pinus massoniana,young Cinnamomum camphora and young Magnolia grandiflora was established for the classification.by taking hyperspectral feature parameters as input eigenvectors and forest species classes as output eigenvectors.The results show that the classification precisions of both mature Pinus massoniana and young Cinnamomum camphora were 100.0% and the composite precisions of the samples were 93.3%.