Temporally Adaptive, Partially Unsupervised Classifiers for Remote Sensing Images

Remote sensing is being increasingly used over the last few decades as a powerful tool for monitoring, study and analysis of the surface of the earth as well as the atmosphere. In this paper we shall consider temporally adaptive pattern recognition techniques for land-cover classification in multitemporal and multispectral remote sensing images. The technique comprises of pre-processing using global and classwise probability density function (PDF) matching for temporally adapting the statistics before classification. We focus on the utility of these techniques in generating improved partially unsupervised land-cover classifiers and their comparative study.