<p>Hyperspectral image classification is challenging to the high dimensionality brought by the huge number of spectral bands for a given spatial region. Classification of the spectral signatures is an approach to identify usage of land or identification of objects like roads, trees, agriculture lands and crops, buildings, etc. It is commonly noticed that hyperspectral images suffer from a labelling dilemma. Methods for classifying huge datasets in high dimensions have been developed more recently. This paper presents a novel approach to classify the hyperspectral imagery using Modified Mutual Nearest Neighbor Clustering (MMNNC). Using the full dimensionality of the pixel data, we compute a similarity matrix. The nearest neighbours are located using the similarity matrix, and dense points collection are discovered for each pixel. These are classified into three categories like noise, weak, and strong points. A homogeneity factor is computed to find clusters for these points. A greedy algorithmic approach is used to label each point into a cluster. The recommended method can produce better classification performance than some of the present methods, according to experimental findings based on numerous hyperspectral image data sets.</p>