论文部分内容阅读
针对传统的光谱角匹配分类算法仅考虑光谱信息,导致混合像元易出现错分和分类结果中出现“麻点”等问题,该文考虑地物连续性这一特点,提出了一种结合像元空间邻域信息对光谱角进行修正的光谱角匹配分类算法。该方法不仅保留了传统光谱角度匹配算法不受增益因素影响和减弱地形对照度影响等优点,并且减小了混合像元被错分的概率。基于ROSIS获取的Pavia大学校园的高光谱影像分类结果表明:加入像元空间邻域信息后的光谱角匹配算法在保证分类精度的同时,有效地减弱了分类结果中的“麻点”现象,验证了该文方法的可行性、有效性。
In view of the traditional spectral angle matching classification algorithm that considers only the spectral information, which leads to the misclassification of mixed pixels and the appearance of “pitting ” in classification results, this paper proposes a Spectral Angle Matching Classification Algorithm Based on Neighborhood Information of Pixel Space to Correct Spectral Angle. This method not only retains the advantages of the conventional spectral angle matching algorithm, which is not affected by gain factors and affects the terrain contrast, and reduces the probability of misclassification of mixed pixels. The result of Hyperspectral Image Classification of Pavia University campus based on ROSIS shows that spectral angle matching algorithm after adding pixel neighborhood information effectively reduces the “pitting” phenomenon in classification results while ensuring classification accuracy , Verify the feasibility and effectiveness of the method.