基于软阈值的高光谱遥感图像分类研究

Classification of Hyperspectral Remote Sensing Images Based on Soft Threshold

对高光谱遥感图像进行分类处理,能够对其中的各种光谱加以高效利用,准确提取地物信息。但是由于分辨率较低,受噪声干扰较严重,导致现有方法对高光谱遥感图像的分类效果不佳,为此,提出基于软阈值的分类方法。为了利于地物提取,方法首先以像元,端元,以及丰度建立L1/2范数模型;然后引入惩罚公式,用于处理由噪声导致的残差;最后分别针对端元,像元,及丰度等参数设计更新公式,并利用目标函数判定其迭代状态,引入交叉验证,对噪声参数与光谱特性采取动态自适应调整。通过实验对比结果,验证提出的软阈值方法具有出色的抗噪声干扰能力,能够更准确的处理光谱差异,有效提升高光谱遥感图像的分类精度。

Classification of hyperspectral remote sensing images can make efficient use of all kinds of spectra and extract the information of ground objects accurately. However, due to the low resolution and serious noise interference, the existing methods are not effective in classification of hyperspectral remote sensing images. Therefore, a classification method based on soft threshold is proposed. In order to facilitate the extraction of terrain objects, the method first establishes a norm model based on pixels, end elements and abundance;then introduces a penalty formula to deal with residual errors caused by noise;finally designs updating formulas for end elements, pixels and abundance parameters respectively, and uses the objective function to determine their iteration state, and introduces cross-validation. The noise parameters and spectral characteristics are adjusted dynamically and adaptively. The experimental results show that the proposed soft threshold method has good performance.