基于可见光/近红外光谱技术的倒伏水稻识别研究  

DISCRIMINATION OF LODGED RICE BASED ON VISIBLE/NEAR INFRARED SPECTROSCOPY

作　　者：刘占宇[1,2] 王大成[1] 李波[1] 黄敬峰 

机构地区：[1]浙江大学农业遥感与信息技术应用研究所,浙江杭州310029 [2]浙江大学环境修复与生态健康教育部重点实验室,浙江杭州310029 [3]浙江省农业遥感与信息技术重点实验室,浙江杭州310029

出　　处：《红外与毫米波学报》2009年第5期342-345,共4页Journal of Infrared and Millimeter Waves

基　　金：国家863计划资助(2006AA10Z203);国家十一五科技支撑项目(2006BAD10A01)

摘　　要：运用可见光/近红外光谱仪获取正常的和受稻飞虱、穗颈瘟危害而倒伏的水稻冠层光谱反射率,采用主成分分析(PCA)方法对反射率光谱进行降维处理,提取2个主分量光谱.其中,第一主分量PC1代表了水稻冠层的光谱特性,第二主分量PC2反映了倒伏水稻的冠层光谱变化信息.将前2个主分量作为支持向量分类机(SVC)的输入向量,建立分类模型.结果表明,对受稻飞虱危害倒伏的水稻验证数据的识别精度为100%,对受穗颈瘟危害倒伏的水稻验证数据的识别精度为90.9%.研究表明可见光/近红外光谱可能是一种有效的倒伏水稻识别方法.Hyperspectral reflectances of the healthy and lodged rice caused by rice planthopper and rice panicle blast were measured with visible/near-infrared （VIS/NIR） spectroradiometer at the canopy level. The principal component analysis （PCA） was used to obtain the principal components （PCs） and to reduce the spectral dimensions of hyperspectral reflectance. Two principal components were extracted. The first （PC1） and second （PC2） reveal the general feature of rice spectral reflectance and spectra change of lodged rice relative to healthy rice,respectively. The front two PCs entered the support vector classification （SVC） as the input vectors to build the discrimination model. The recognition accuracies of healthy and lodged rice are 100% and 90.9% for the rice planthopper and rice panicle blast stresses,respectively. The results demonstrate that visible/near-infrared spectroscopy technique may provide potential discrimination accuracy for lodged rice.

关 键 词：稻飞虱 穗颈瘟 可见光/近红外光谱反射率 主成分分析 支持向量分类机 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] S43[自动化与计算机技术—控制科学与工程]