Feature detector for space hyperspectral images

Subject of study. This study focuses on methods of feature detection applied to the processing of multichannel hyperspectral images obtained from space-based devices, particularly those based on optical channel splitting. Aim of study. The aim was to develop or modify an existing feature detector according to the constraints of hyperspectral imaging, including variations in sensor types, absolute intensity levels, and signal-to-noise ratios across different channels. Method. The approach leverages visible and infrared spectral information to improve feature detection algorithms that traditionally rely solely on panchromatic data. Main results. The Oriented FAST and Rotated BRIEF detector was identified, owing to its high image processing speed, through a literature review. However, applying this algorithm to single spectral channel images (converted to grayscale) proved challenging because of low contrast, which reduced the detection rate of strong features. The number of detected features was increased by integrating spectral information into the feature detector. The algorithm was tested on Hyperion dataset samples, and the results were compared with those obtained using the conventional Oriented FAST and Rotated BRIEF detector paired with histogram equalization and Oriented FAST and Rotated BRIEF with a lowered detection threshold. The proposed algorithm demonstrated an average accuracy improvement of 11% and a 20% increase in processing speed. Practical significance. The proposed method provides a foundation for stitching hyperspectral images captured by a camera with split optical channels for visible and infrared ranges, such as those deployed on the International Space Station.