Abstract
In this paper, we propose an algorithm named SS_DPC for hyperspectral image classification. First, the image is segmented into hyperpixels according to spatial and spectral information, which are used as basic units for clustering instead of pixels. Computing the inner product of the local density and the minimum inter_cluster distance for each unit, Density peaks clustering (DPC) algorithm sorts products in descending order and selects the globally optimal solutions as cluster centers. The following conclusions have been verified through experiments:(1)Proper quantity of superpixel (K value) can improve the consistency between clustering results and actual values effectively.(2)Image segmentation can weaken the interference of abnormal data, so the ARI values of SS_DPC, SS_K_Means are higher than that of K_Means significantly.(3)SS_DPC algorithm is much better than other clustering algorithms in precision and robustness.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have