独立分量分析与主成分分析方法的湿地遥感分类精度对比——以西洞庭湖湿地为例

Comparison of Accuracy on Wetland Remote Sensing Classification between Independent Component Analysis and Principal Component Analysis Methods——A Case Study of Wetlands in Western Dongting Lake

拟探索提高中、高分辨率遥感影像湿地分类精度的新方法,为洞庭湖湿地研究提供方法支持。通过将独立分量分析方法应用于遥感图像分类,并与主成分分析方法结果进行对比,验证其是否能有效提高湿地分类精度。结果表明,应用独立分量分析和主成分分析方法处理遥感影像,没有造成图像信息损失;Landsat5TM影像经过两种方法处理后,影像清晰度变差,但并不足以影响典型湿地类型的目视判读;SPOT5影像经过两种方法处理后,影像更加清晰。独立分量分析方法可以明显提高典型湿地类型的可分性,但对于草滩地和水田的分类仍然存在缺陷。Landsat 5TM影像经独立分量分析算法处理后,总体精度比原始影像提高11.83%,比用主成分分析方法处理后的影像精度高5.35%;SPOT 5影像经独立分量分析算法处理后,总体精度比原始影像提高10.7%,比用主成分分析方法处理后的影像精度高5.07%。独立分量分析基于高阶统计信息,不但能去除波段之间的相关性,而且可以得到分量之间相互独立的特性,增强不同湿地类型的可分离性,从而提高了信息提取的精度。

This study aimed at exploring method to improve the classification accuracy of high resolution remote sensing image, which could support the research on the wetlands in Dongting Lake area. The results showed the information of images processed by independent component analysis （ICA） and principal component analysis （PCA） methods had not been lost; after PCA and ICA processing, the sharpness of the images were deteriorated, but not enough to affect the visual interpretation of the typical wetland types; SPOT 5 image became clearer after PCA and/CA processing. ICA method could significantly improve the divisibility of the typical wetlands in Dongting Lake, but it still has shortages on classify marshes and paddy fields. For Landsat 5 TM image, compared to the original image, the overall accuracy increased by 11.83% after the image processed by ICA method, that was 5.35% higher than by PCA method. For SPOT 5 image, compared to the original image, the overall accuracy increased by 10.7% after the image processed by ICA method, that was 5.07% higher than by PCA method. Based on the higher-order statistics information, independent component analysis method could not only remove the correlation between the bands, but also obtain the independent component characteristics, enhancing separability between wetland types. It also could effectively remove the negative impact of the typical wetland classification, and improve the accuracy of wetland information extraction.