Extraction Technology of Tree Species Information of Hyperspectral Remote Sensing Based on Improved BPNN | 基于改进BP神经网络的高光谱遥感树种信息提取技术

Китайский. 【目的】通过分析树种间的光谱差异及改进分类算法以提高树种信息提取精度。【方法】文章采用安徽省砀山县EO-1Hyperion影像，通过不同树种光谱信息的差异分析，筛选出区分树种信息的光谱指标，并采用改进的BP神经网络模型完成树种信息提取。【结果】结果表明，原始反射率和一阶微分部分光谱波段可用于树种识别，且一阶微分光谱的差异大于原始反射率；引入动量项和遗传算法改进的BP神经网络模型树种识别精度较传统BP神经网络提高8．5％，Kappa系数提高0．12。【结论】该方法可以实现较为准确的树种信息提取，能够达到对林业工程进行监测的目的，对快速评价工程质量有重要意义。

Английский. [Objective] Our aim was to enhance the accuracy of species information extraction by analyzing spectral differences between species and improving classification algorithms. [Method] EO-1 Hyperion images of Dangshan in Anhui were selected in this study and spectral indexes that can distinguish tree species information were filtered by analyzing tree species spectrum. After-wards, information of tree species was extracted using the improved BP neural network model. [Results] The results showed that original reflectivity and part of first order differential spectral bands could be used to identify tree species and the difference of first order differential spectrum was larger than that of original reflectivity. The identification accuracy of tree species from the improved BP neural network model （add momentum and genetic algorithm） was increased by 8.5% compared to that of traditional BPNN and Kappa coefficient was enhanced by 0.12. [Conclusion] The results noted in this study could achieve a more precise information extraction of tree species and meet demand of forestry project monitoring. This method has important implication for the rapid assessment of project quality.