基于提升算法的超谱遥感图像融合分类研究

超谱遥感技术的发展对遥感图像处理算法提出了新的挑战,超谱遥感图像所特有的高光谱维数,使适用于多光谱图像的算法不适合直接用于超谱图像.利用数据融合技术可以将超谱图像从高维降到低维,因而有利于图像的分析和处理.提升算法是构造第2代小波的关键技术,该文研究了其用于超谱遥感图像融合分类的可行性,利用提升算法将第1代小波改造成第2代小波,并对标准的AVIRIS超谱遥感图像实现图像融合,在融合的同时,提取图像的光谱特征用于分类,在相同的实验标准下在像素层和特征层上分别对图像进行了第2代小波融合分类,并用分类精度对实验结果进行了客观的评价.实验结果表明,以提升算法构造的特征层小波融合分类比像素层分类精度提高了7.78%.

Fusion classification research of hyperspectral remote sensing image based on lifting scheme

The development of hyperspectral remote sensing technology has presented new challenges to the algorithms of processing remote sensing images.The high dimensions of hyperspectral remote sensing images disable the algorithms of multispectral images to be used directly in hyperspectral image.However,the data fusion technique can decrease high dimensions to low dimensions so that analysis and processing of a hyperspectral image are simplified.The lifting scheme is a key technique in constructing the second generation wavelet, whose feasibility in classification of a hyperspectral remote sensing image was researched. A second generation wavelet was constructed by using the lifting scheme onto a first generation wavelet, then implementing it on standard AVIRIS hyperspectral remote sensing image fusion, at the same time, extracting spectral features in order to classify the hyperspectral image.Under the same experiment circumstance,a second generation wavelet was implemented on pixel and feature level fusion classification of hyperspectral image. The result was evaluated objectively by using classification accuracy. Experiments show that second generation wavelet fusion classification accuracy of feature level is 7.78% higher than that of pixel level.