基于遥感影像的神经网络立地质量评价研究  

Study on forest site evaluation with artificial neural network based on remote sensing image

作　　者：巩垠熙[1,2] 高原[1] 仇琪[1] 谢飞 冯仲科[1] 樊江川[1] 

机构地区：[1]北京林业大学测绘与3S技术中心,北京100083 [2]国家测绘局第一航测遥感院,陕西西安710000 [3]国家测绘地理信息第一地理信息制图院,陕西西安710054

出　　处：《中南林业科技大学学报》2013年第10期42-47,52,共7页Journal of Central South University of Forestry & Technology

基　　金：国家科技支撑计划项目(2012B AH34B01);国家自然科学基金(30872038)

摘　　要：传统立地质量评价体系主要使用地面调查数据,而多光谱遥感影像为大范围研究森林生产力和立地状况提供了便捷的途径。为了改进传统的立地质量评价体系,以内蒙古旺业甸林场为研究对象,基于研究区域的遥感影像结合地表小班调查数据,采用BP人工神经网络模型,以落叶松为例,建立了遥感因子结合立地因子与地位指数关系的神经网络模型,进行立地质量评价研究。建立了BP人工神经网络地位指数预测模型,模型的预测精度达到90.97%,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比较,精度提高了5.44%。结果表明多光谱遥感影像十分适用于森林立地质量评价,充分证实了本研究方法的有效性和优越性。The classic site quality evaluation system merely adopts artificial ground survey data, while multispectral remote sensing （RS） data provide a convenient way for large-scale studies on forest productivity and site quality. In order to improve the classic system, a neural network model which combined RS factors with site factors and site index relations was established and used to study the sub- lot site quality evaluation in the Wangyedian Forest Farm in Inner Mongolia Autonomous Region. Based on back propagation artificial neural network （BPANN）, this model combined multispectral RS data with sub-lot survey data, and took Larch as example. The accuracy of BPANN site index prediction model went up to 90.97%, which was 5.44% higher than that of the neural network model based on classic sub-lot survey data. The results indicate the applicability of multispectral RS data in forest site quality evaluation, and fully confirm the effectiveness and superiority of this new method.

关 键 词：遥感影像 神经网络 立地质量 植被指数 

分 类 号：S712[农业科学—林学]