Texture based feature extraction: Application to burn scar detection in Earth observation satellite sensor imagery

Abstract

A single band texture-based burn scar identification algorithm incorporating the use of grey level co-occurrence matrices with a low pass filtering technique is described and demonstrated using 1 km resolution ATSR-2 imagery of burned savannas in southern Sudan. The algorithm results are compared to those produced by the iterative intensity-based isodata classification technique. The accuracy of each of these methods was evaluated by comparison with 18 m spatial resolution imagery. For a set of 22 sample fire scars of varying area Pearson correlation coefficients of 0.75 and 0.94 were obtained between the burnt area statistics produced with the low-spatial resolution texture and isodata methods respectively and those produced using the high-resolution data. The classification quality, as described by the Kappa (k) statistic, produced values of kTEXTURE = 0.558 and kISODATA = 0.852. Texture is shown to be an image variable capable of highlighting burned area in low spatial resolution imagery, but the currently tested approach offers no accuracy of quality benefit over the solely intensity-based method.