Abstract: The spectral similarity measure is the key to extract the information from hyperspectral remote sensing imagery. A new changing-weight spectral similarity, called Spectral Changing-Weight Similarity Measure (SCWM), was proposed based on the combination of Euclidean distance and spectral angle cosine. According to different land covers, SCWM can automatically alter the weight of Euclidean distance and spectral angle cosine. The classification accuracy of different spectral similarity measures is compared, using the misclassification rate of the standard spectral library data and the confusion matrix of the airborne OMIS hyperspectral image. The experimental results demonstrate that Spectral Changing-Weight Similarity Measure is more effective than the spectral similarity measure by taking into account one spectral feature or two spectral features to the precise classification.

Key words: hyperspectral image, similarity measure, Spectral Changing-Weight Similarity Measure, remote sensing classification