基于多角度融合的CHRIS数据提取湿地植被的研究

Using multi-angle hyperspectral data, the spectral characteristics of typical vegetation communities in the East Dongting Lake wetland at different angles are analyzed, the best method for multi-angle information fusion is determined, and the wetland vegetation types are finely identified from the fused images. In this paper, CHRIS multi-angle hyperspectral data is used to study and calculate the best band combination and angle for narrow-band NDVI, and to evaluate the pixel-level fusion method of CHRIS 0° image and NDVI. The calculation results show that the best red band and near-infrared band for NDVI are located at 667.6 nm and 926.95 nm respectively, relative to the 24th band and 55th band of the CHRIS data. By selecting four fusion methods, namely HSV, Brovery, Gram - Schmidt and PCA, it can be known that the PCA fused image has the least loss of spectral information, richer texture details and the largest amount of information. The overall accuracy of the PCA fused image is 81.36%, which is 7.93% higher than that of the single-angle image fusion, and the Kappa coefficient is increased by 0.0976, significantly improving the omission error of Carex and the commission error of Artemisia selengensis. Therefore, the multi-angle information fusion based on NDVI is an effective way for fine vegetation identification. Multi-angle information fusion enriches the amount of information of ground objects and improves the identification accuracy of ground objects.

利用多角度高光谱数据，分析不同角度下东洞庭湖湿地典型植被群落的光谱特征，确定多角度信息融合的最佳方法，并对融合影像进行湿地植被类型精细识别。本文使用CHRIS多角度高光谱数据，研究计算窄波段NDVI的最佳波段组合和角度，评价CHRIS 0°影像与NDVI的像素级融合方法。计算结果显示：NDVI的最佳红波段和近红外波段分别位于667.6 nm和926.95nm，相对于CHRIS数据的第24波段和第55波段。选取HSV、Brovery、Gram-Schmidt和PCA四种融合方法可知：PCA融合图像的光谱信息丢失最少、纹理细节更丰富，信息量最大。PCA融合影像总体精度81. 36%，比单一角度影像融合精度提高7.93%，Kappa 系数提高,0。0976，使得苔草的漏分误差和泥蒿的错分误差得到明显改善。因此，基于NDVI的多角度信息融合是植被精细识别的一种有效途径，多角度信息融合丰富了地物的信息量，提高地物识别精度。