Role of Individual Samples in Modified Possibilistic c-Means Classifier for Handling Heterogeneity Within Mustard Crop

Abstract

In raster remote sensing images within class have variations represented as heterogeneity. Pixel-based classifiers use means/variance-covariance (DVC) statistical parameters, generated from training sample datasets. These parameters do not represent in totality about variations within class. This research paper explains the role of each sample in handling heterogeneity without using statistical parameters from the training samples. Modified possibilistic c-means fuzzy algorithm capable of mapping single class to handle heterogeneity has been experimented. Multi-spectral temporal images of Sentinel-2A/B of Banasthali, Rajasthan region acquired from 1 November 2019 to 24 February 2020 have been used for mustard class mapping. It has been observed that while using individual samples in place of statistical parameters in fuzzy-based classifiers, individual class identified has been least affected due to heterogeneity within class.