基于多角度融合的CHRIS数据提取湿地植被的研究

目的]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。方法]使用CHRIS多角度高光谱数据,针对洞庭湖湿地植被的光谱特征,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,进而对洞庭湖地区湿地植被进行提取。结果]计算NDVI的最佳红波段和近红外波段分别位于667.6 nm和926.95 nm,对应于CHRIS数据的第24波段和第55波段;选取HSV、Brovery、Gram-Schmidt和PCA 4种融合方法进行融合,发现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融合影像的总体精度为81.36%,比单角度影像提高7.93%,Kappa系数提高0.097 6,且苔草的漏分误差和泥蒿的错分误差得到明显改善。结论]基于NDVI的多角度信息融合是提高湿地植被识别精度的一种有效途径,多角度信息融合丰富了地物的信息量,提高地物识别精度。

Research on the Extraction of Wetland Vegetation Information from CHRIS-PROBA Data Based on Multi-angle Image Fusion

Objective]The purpose of this paper is to analyze the spectral characteristics of typical vegetation communities in East Dongting Lake Wetland using spaceborne angular hyperspectral imagery and research on the best fusion method based on multi angle information to identify the wetland vegetation types precisely. Method]According to the spectral feature of wetland vegetation in Dongting Lake, the optimal angle and the best band combination were studied to calculate narrow-band NDVI using CHRIS data. It focuses on evaluating the methods of fusing 0 degree image and NDVI in pixel level. And then wetland vegetation information in Dongting Lake was extracted. Result]The results show that the best red band and near infrared band to calculate NDVI locate at 667.6 nm and 926.95 nm, corresponding to the 24th and 55th band of CHRIS data. HSV, Brovery, Gram-Schmidt and PCA fusion methods were selected to evaluate the fusion effect. It is found that the image used PCA fusion method suffered from the least loss of spectral information and had the richest texture details and maximum information content. The overall classification accuracy of multi-angle fusion image is 81.36%, 7.93% higher than that of single angle image. The Kappa coefficient improved 0.097 6. In addition, the omission error of Carex and the misclassification error of Artemisia selengensis got obvious improvement. Conclusion]It shows that multi-angle fusion based on NDVI is an effective way to improve the accuracy extracting vegetation information. Multi-angle information fusion can enrich the information of the observation target and improve the accuracy of recognition accuracy.