Study of the Clustering Algorithms for Hyper Spectral Remote Sensing Images

Abstract

The data taken from the hyperspectral images are discrete and hard to classify because they are arranged in the contiguous spectral bands. We can easily detect and classify the data from the spectral images if the number of attributes in the images is very little. But it is very difficult to segregate the data from the images if the numbers of classes are more. To make the segregation easy we implement the procedure that utilizes a clustering algorithm. This paper comprises of two sections, firstly to perform unsupervised learning using different types of clustering algorithms and secondly, to compare the efficiency of the resultant clustering of these different methods to prove that which clustering method is best suitable in reading the hyperspectral imaging data. For this I have used these clustering algorithms, they are DBSCAN, MiniBatch K-Means, K-Means. By comparing these techniques I surmised that the K-Means is better for using the HyperSpectral Imaging data. To perform these calculations I used the Matlab data set from the Computational Intelligence Group.