Segmentation model for hyperspectral remote sensing images based on spectral angle constrained active contour

We propose an alternative hyperspectral image segmentation method based on a geometric active contour model. First, we use a spectral angle as a measure index to measure the spectral similarity between the pixels, and the spectral angle constrained function is constructed based on spectral similarity. Then, depending on the class separability criterion, an optimal band is chosen which is suitable for image segmentation. The model retains advantages of the traditional C–V model in regional information with rich textures and edges. Its unique feature is exploiting the characteristics of hyperspectral remote sensing images in regions with abundant spectral information. As a consequence, the capture capability, segmentation speed, and accuracy are enhanced. Finally, we present experiments using WorldView and Hyperion images, from whose results demonstrate the improved attributes of the proposed method over those of the traditional C–V model. The proposed model provides better results than the traditional models both in segmentation accuracy and computational efficiency.