光谱信息降维及判别模型建立用于识别湿地植物物种

Reduction of Hyperspectral Dimensions and Construction of Discriminating Models for Identifying Wetland Plant Species

研究分析了北京地区常见湿地植物的高光谱数据特征。主要采用马氏距离法和主成分分析法对光谱进行降维,并对光谱特征进行分析和提取;利用提取的光谱信息构建判别模型对湿地植物进行判别,并对模型精度进行比较评价,最后获得最佳判别模型。研究结果显示:(1)马氏距离法和主成分分析法都能对光谱进行有效降维,(2)利用从光谱中提取的特征参数建立的判别模型,得到物种识别的精度都高于90%。可见,湿地植物高光谱数据信息的变换和利用能促进对湿地植物光谱特征的认识和提取,而建立模型进行植物判别可以指导未来遥感领域的湿地制图和监测。

The present paper researched and analyzed the hyperspectral data of wetland plant species often occurred in Beijing.The methods of Mahalanobis Distance（MD） and principal component analysis（PCA） were mainly applied to reduce the dimensions of hyperspectral data and to analyze and extract the features of spectra.The authors use the extracted spectra to build identification models for identifying the wetland species.The authors then compared and evaluated the precisions of models and finally obtained the best discriminating model.The results showed that（1） the dimensions of hyperspectral data can be efficiently reduced by both MD and PCA methods.（2） The discriminating models established using the parameters extracted from the resulting spectra of MD and PCA could identify the wetland plants with high precisions of more than 90%.As a result,the conversion and usage of the hyperspectral data can help better understand and well extract the spectra of different wetland plants.Furthermore,the constructed discriminating models for wetland species could also be used in the future to guide us in mapping and monitoring of wetland ecosystem by applying the remote sensing data.