ABBD: Accumulated Band-Wise Binary Distancing for Unsupervised Parameter-Free Hyperspectral Change Detection

Abstract

As a fundamental task in remote sensing earth observation, hyperspectral change detection (HCD) aims to identify the changed pixels in bitemporal hyperspectral images. However, the water-absorption effect, poor weather conditions, noise and inconsistent illumination as well as lack of accurate ground truth has made HCD particularly challenging. To tackle these challenges, a novel Accumulated Band-wise Binary Distancing (ABBD) model was proposed for unsupervised parameter-free HCD. Rather than relying on the absolute pixel difference with thresholding in conventional approaches, the binary distancing only indicated whether a pixel was changed or not in a certain band, which could alleviate the adverse effects of noise-induced inconsistency of measurement. The band-wise binary distance map is then accumulated to form a grayscale change map, on which the simple k-means was applied for a final binary decision-making. Experiments on three publicly available datasets have validated the superiority of our approach, which has yielded comparable or slightly better results in comparison to a few state-of-the-art methods including several deep learning models.