Semantic-Aware Image Filtering for Classification of Hyperspectral Images

The success of spectral-spatial hyperspectral image classification techniques are dependent on their ability to consider appropriate spatial information. Structure preserving image filtering technique preserves structures of the objects while removing the noises from the image. To take into account better spectral-spatial information, in this research we have constructed a profile that consists of multiple filter images generated by applying a recently developed semantic-aware image filtering technique. Then, the pixels on the profile represented with spectral-spatial features are used to classify the hyperspectral images. The experiments conducted on three real hyperspectral datasets confirm potentiality of the proposed technique.