EVALUATION OF THE LINEAR MODEL TECHNIQUE OF SPECTRAL MIXING AS A SUBSID TO THE CLASSIFICATION OF LAND USE AND OCCUPATION

Abstract

The development of remote sensing technologies has encouraged the application of digital images in various circumstances, including environmental studies and Earth monitoring. For classification studies of land use and occupation, several classification techniques can be applied. These, however, vary depending on the characteristics of the imaging sensors and the purpose of the experiment. In image classification studies, it is common to encounter problems of spectral mixing that can be limiting, so working methods are used to extract information from images with greater detail, considering the properties of the materials present within a pixel and to help in classification techniques. This work approaches the linear model of spectral mixture as subsidy to the classification of land use and occupation. The results were compared with the classification performed from Sentinel 2-A satellite images. The overall accuracy for the ratings sourced from the Sentinel-2A and LMSM data was 65% and 62%, respectively, and the Kappa coefficient was 0.53 (satellite spectral data) and 0.62 (LMSM result) indicating that the quality of the two mappings can be considered as moderate, good enough or good.

Keywords: Spectral mixing. Linear spectral mixing model. Image classification. Sentinel-2A.