Mathematical methods for remote sensing -- a review (Q2756195)

The spatial and temporal resolution of space derived remotely sensed data has improved dramatically in the, last decade and is soon set to rival aerial data. A variety of mathematical techniques have the potential for use in the pre-processing and analysis of this data. The scenario today is that only a very small set of these tools is being used. The objective of this paper is to identify these techniques which have shown the potential for this purpose. NEWLINENEWLINENEWLINEThe topics addressed are Image Restoration, Orbit/altitude modelling, Fuzzy classification, Neural Network models, Mathematical morphology, Knowledge based Systems, Wavelet transforms and Fractals. The higher spatial resolution and complex viewing geometries etc. call for the application of image restoration algorithms, especially those where the point spread function is not known precisely. The need for ground control points has limited the accuracy of position obtainable. By adopting better orbit/altitude modelling, the required number of GCPs can be drastically reduced. The strait jacket decision boundaries of Bayes' classifier can give way to more meaningful `fuzzy' boundaries if fuzzy techniques are adopted. Similarly, the learning capabilities of ANNs enable higher accuracies of classification. The higher spatial resolution with its impact on shape call for shape sensitive methods like mathematical morphology. The higher specificity of wavelets compared to Fourier transforms also is helpful in data compression and retrieval especially for digital libraries. Fractals again provide a classification method as well as excellent data compression and texture analysis capabilities. The paper also identifies certain future activities like pyramidal processing, feedback ANNs, GIS modelling etc. Finally, a tentative syllabus is for short courses which could enhance the applicability of modern mathematical tools for remote sensing data analysis is also included.