摘要
多光谱遥感数据蕴含着大量的地表立地信息,而传统立地质量评价体系主要使用了人工地面调查数据。为了建立一套有效的立地质量评价体系,以内蒙古赤峰市旺业甸林场为研究对象,基于研究区域的多光谱遥感数据结合地面小班调查数据,采用一种改进的反向传播人工神经网络(back Propagation artificial neural network,BPANN)模型,以落叶松为例,建立了遥感光谱因子结合立地因子与地位指数关系的神经网络模型,对研究区域的小班进行立地质量评价研究。通过训练数据集的敏感度分析剔除弱相关或不相关的因子,简化了神经网络的规模,提高了网络的训练效率,得到了最优的地位指数预测模型,模型的预测精度达到95.36%,与使用传统小班调查数据建立的神经网络模型的预测结果进行了比较,精度提高了9.83%,说明使用多光谱遥感数据+小班调查数据确定的落叶松地位指数预测模型具有最高的预测精度。多光谱遥感数据十分适用于森林立地质量评价,改进BP神经网络具有理想的预测精度,充分证实了该方法的有效性和优越性。
Multispectral remote sensing data containing rich site information are not fully used by the classic site quality evalua- tion system, as it merely adopts artificial ground survey data. In order to establish a more effective site quality evaluation sys- tem, a neural network model whieh combined remote sensing spectra factors with site factors and site index relations was estab- lished and used to study the sublot site quality evaluation in the Wangyedian Forest Farm in Inner Mongolia Province, Chifeng City. Based on the improved back propagation artificial neural network （BPANN）, this model combined multispectral remote sensing data with sublot survey data, and took larch as example, Through training data set sensitivity analysis weak or irrelevant factor was excluded, the size of neural network was simplified, and the efficiency of network training was improved. This opti- mal site index prediction model had an accuracy up to 95.36%, which was 9.83% higher than that of the neural network model based on classic sublot survey data, and this shows that using multi-spectral remote sensing and small class survey data to deter- mine the status of larch index prediction model has the highest predictive accuracy. The results fully indicate the effectiveness and superiority of this method.
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2013年第10期2815-2822,共8页
Spectroscopy and Spectral Analysis
基金
国家科技支撑计划项目(2012BAH34B01)
北京市自然科学基金项目(09D0297)
国家自然科学基金项目(30872038)资助
关键词
立地质量
多光谱遥感
神经网络
Site quality
Multispectral remote sensing
Neural Network
