基于PCA和PNN的水稻病虫害高光谱识别  

Hyperspectral identification of rice diseases and pests based on principal component analysis and probabilistic neural network

作　　者：李波[1,2] 刘占宇[2,3,4] 黄敬峰[2,3,4] 张莉丽[5] 周湾[5] 石晶晶[4] 

机构地区：[1]浙江大学公共管理学院,杭州310028 [2]浙江省农业遥感与信息技术重点实验室,杭州310029 [3]浙江大学农业信息科学与技术中心,杭州310029 [4]浙江大学农业遥感与信息技术应用研究所,杭州310029 [5]杭州市植保土肥总站,杭州310020

出　　处：《农业工程学报》2009年第9期143-147,共5页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家863计划资助项目(2006AA10Z203);国家科技支撑项目(2006BAD10A01)

摘　　要：对水稻病虫害准确、快速的识别是采取病虫害防治措施的基础,同时对灾害评估也具有积极意义。该研究选用在水稻孕穗期时测定的两期受稻干尖线虫病危害的水稻叶片光谱数据和于水稻分蘖期时测定的两期受稻纵卷叶螟危害的水稻叶片光谱数据,通过对水稻叶片的光谱特征分析,选用可见光波段(490～670nm)和短波红外波段(1520～1750nm),用主成分分析技术(PCA)对上述光谱波段进行压缩,获得主分量光谱,最后结合概率神经网络(PNN)对稻干尖线虫病和稻纵卷叶螟进行识别,结果显示对水稻病虫害的识别精度高达95.65%。研究表明,PCA和PNN相结合,可以实现对多种水稻病虫害进行快速、精确的分类识别。Correct and fast identification of rice diseases and pests was the basis of diseases and pests prevention measures,and significant in disaster assessment.This study adopted spectral reflectance of rice leaves stressed by rice Aphelenchoides besseyi Christie of two periods at the rice booting stage and by rice leaf roller of two periods at the rice tillering stage.With the analysis of the spectral characteristics of rice leaves,visible band （490-670nm） and short wave infrared band （1520-1750nm） were selected with principal component analysis （PCA） transformed from the above two selected band. The recognition precision of rice Aphelenchoides besseyi Christie and rice leaf roller using probabilistic neural network （PNN） was as high as 95.65%. The research demonstrated that the method was feasible and reliable to precisely identify non-healthy rice stressed by rice diseases and pests from healthy rice based on PCA and PNN.

关 键 词：光谱分析 主成分分析 神经网络 水稻 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] S435.11[自动化与计算机技术—控制科学与工程]