Accuracy Assessment and Validation of Large-Scale Subpixel Classified Maps for Remotely Sensed Data

Classification is one of an important approaches used in remote sensing application. The accuracy assessment of classified dataset is one of the major concerns while considering large-scale region. While dealing with large areas like undulating Himalayas where topography is varying at each point, we have to consider each pixel existing over there in the validation process to generate more accurate results. In the present paper, an approach is presented to validate the classified dataset at each and every pixel level using normalized difference index. The case study has been conducted over a part of Bada Shigri, the largest glacier in Himachal Pradesh, India. To implement the proposed work, data is acquired through Landsat-8 as primary dataset and Sentinel-2 as a reference dataset. The linear spectral mixing (LSM) as subpixel classifier has been used to classify the Landsat-8 dataset at subpixel level, and validation of the classified outputs at each pixel level is done by using NDSI maps generated from HR Sentinel-2 data. Accuracy assessment procedure has been followed for all the available pixels in the imagery. From experimental outcomes, it is concluded that the normalized difference index plays a significant role in accurate validation of large-scale subpixel classified maps for remotely sensed data.