| Summary: | Unlike conventional images, which have three channels of information, hyperspectral images are composed of many spectral channels that provide detailed information about the materials present in them. Thus, considering their great potential to monitor changes in the environment and the importance of freshwater bodies for life and nature, it is relevant to propose and evaluate the effectiveness of different computational methods focused on detecting bodies of water in hyperspectral images; therefore, this research proposes and evaluates a computational method based on Fourier phase similarity. To do so, four methodological phases were defined, namely: exploration and selection of open-source technologies for hyperspectral image analysis, determination of the characteristic pixel of water bodies, calculation of Fourier phase similarity between the representative pixel of water bodies and the 200 sample pixels chosen from water bodies and other materials, and verification of the method on a test hyperspectral image. Spectral, NumPy, and Pandas libraries of Python were used to implement the proposed method, which resulted, for the first 170 bands, on an average phase similarity of 99.46% with respect to water body pixels and a minimum phase similarity with water body pixels of 93.01%. The results show that the proposed method is effective to detect water body pixels and can be used or extrapolated as an alternative to detection methods based on correlation metrics and machine learning.
|
|---|
