Abstract: In order to obtain the best identification parameters of the main plants in the desert grassland of Seriphidium transillense,images of community were collected by SOC710 VP imaging spectrometer in April,Seriphidium transillense, Ceratocarpus arenarius,Petrosimonia sibirica and bare land were extracted as recognition objects,Identification parameters were selected based on spectral response and peak-valley characteristics,including eight position parameters,two area parameters and four vegetation indexes,the sensitive parameters were selected one by one according to the significance difference of variance analysis,and the fisher discriminant model was used for accuracy verification. The results show that:The reflectance of four kinds of recognition objects showed different characteristics in visible and near infrared band;The Red Valley position,red edge position,Red Valley amplitude,blue edge area,NDVI₁,RVI₁ and RVI₂ were used for discrimination,the discrimination accuracy of the four objects was 91.11% for S. transiliense,80.56% for C. arenarius,91.11% for P. sibirica,100% for bare land,and 92.13% for total accuracy. Therefore,a discriminant model based on these identification parameters is useful for further quantitative species classification of community images.

Key words: Grassland plants, Spectral characteristics, Identification parameters, Screening, Fisher discriminant

摘要： 为获得伊犁绢蒿荒漠草地主要植物最佳识别参数，本研究利用SOC710 VP成像光谱仪于4月采集群落影像，以伊犁绢蒿（Seriphidium transiliense）、角果藜（Ceratocarpus arenarius）、叉毛蓬（Petrosimonia sibirica）和裸地为识别对象，基于光谱学响应与峰谷特性选取8个位置参数、2个面积参数、4个植被指数，按照显著性差异从小到大逐一筛选敏感参数，并使用Fisher判别分析进行精度验证。结果表明：4种识别对象的反射率大小在可见光和近红外波段表现出不同的特征；用筛选出的红谷位置、红边位置、红谷幅值、蓝边面积、NDVI₁、RVI₁和RVI₂进行判别，精度分别为伊犁绢蒿91.11%、角果藜80.56%、叉毛蓬91.11%、裸地100%，总精度为92.13%。本试验筛选出的识别参数建立判别模型可为进一步对群落影像进行物种定量分类提供依据。

关键词: 草地植物, 光谱特征, 识别参数, 筛选, Fisher判别