Cross-diffusion based filtering as pre-processing step for remote sensing procedures. (February 2020)

• Record Type: Journal Article
• Authors: Cuesta, Eduardo; Quintano, Carmen; Fernández–Manso, Alfonso
• Is Part Of: Advances in engineering software. Volume 140(2020)
• Publication Date: 2020-02
• Subjects: Cross-diffusion systems -- Image filtering -- Remote sensing; 68U10 -- 97M50; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2019.102751
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: Nonlinear cross-diffusion based filtering of satellite images of damaged areas prior their classification as affected/non-affected is tested. Classification of filtered images outperformed classification of original images (i.e. its accuracy was higher). Several vegetation indexes and synthetic bands acted as inputs to the filter and/or classifier. Abstract: A new methodology combining 2 × 2 cross-diffusion systems of nonlinear partial differential equations (CDS) with classical image classification procedures is proposed in the present paper. Such a kind of mathematical models (CDS) have been theoretically studied in previous works in the context of image processing, however here they are tested and stressed in very practical instances. In particular, the main contribution of this paper is the improvement of the classification of satellite images when they are previously filtered by means of a CDS model. This conclusion is based on a wide and costly experimentation with satellite images of areas damaged by forest fires and surface coal mining, all of them located in Mediterranean areas. The efficiency of our methodology is not only in terms of the classification improvement but also in terms of the runtime saving since CDS based filtering is much less costly than other classical partial differential equations based filtering mathematical models as for example anisotropic models or higher order ones, always within the framework of nonlinear partial differential equations.