Proposal of a computational method for asbestos detection in hyperspectral images based on the identification of prominent peaks in the spectral signature

This study proposes a computational method for asbestos detection in hyperspectral images. The methodology consists of five phases: selection of sample pixels and identification of the characteristic pixel, determination of prominent peaks in the spectral curve, method implementation with reference threshold definition, application to test images, and comparative evaluation of effectiveness and efficiency. The method identifies asbestos pixels by calculating the Euclidean distance between the prominent peaks of spectral curves. Results show no overlap between maximum distances of asbestos pixels and minimum distances of non-asbestos pixels, detecting 11.87% of asbestos pixels in the test image. Although the correlation method is 1.02% faster, the difference is negligible. This method can be extrapolated to other materials with similar spectral features, contributing to urban diagnostics of hazardous materials like asbestos.