Band selection methods impact city hyperspectral image segmentation performance
Band-Selection Stability and Semantic Segmentation Performance: A Study on Hyperspectral City
Computer Vision and Pattern Recognition
Summary
High-dimensional hyperspectral images have many color bands, which makes processing them hard. The authors compared six ways to pick important color bands from these images and checked how stable these methods are when picking bands on different samples. They tested how well these smaller sets of bands work with models that label city scenes and found that some methods are more consistent in picking similar bands but this consistency doesn't always mean better labeling results. They also showed that using fewer bands can make the models much faster while keeping good accuracy.
What this means in practice
- •For autonomous vehicle developers: Use stable band-selection methods to reduce hyperspectral data size, speeding up real-time city scene understanding without losing accuracy.
- •For remote sensing analysts: Improve land cover segmentation of urban regions with fewer spectral bands, allowing faster processing for monitoring and mapping.
Tested on one dataset.
Authors
Jiarong Li, Imad Ali Shah, Enda Ward, Martin Glavin, Edward Jones, Brian Deegan
Abstract
Resource constraints make high-dimensional hyperspectral imaging challenging in autonomous perception, motivating the use of band selection methods. However, the sensitivity of band-selection methods to sampled data and their relationship to semantic segmentation models (SSMs) remain underexplored. This study evaluates six band selection methods on ten independently sampled, class-balanced region-of-interest (ROI) sets, yielding 60 top-25 band subsets from the Hyperspectral City V2 (128 bands: 450-950nm) dataset. Top-$K$ bands ($K\in\{3,5, ... 13\}$) from the first three ROI sets are evaluated with three SSMs against the corresponding 128-band baseline. Experiments show that intra-method stability is method-dependent: Sim-LP shows the highest stability (pairwise Jaccard similarity) and, together with JMIM+CSNR, yields the best segmentation results. Top-$K$ based SSMs remain competitive with baselines, with gains of up to 2.01 mIoU and 1.72 mF1 points, and 18-22x faster CPU inference for $K=9$. However, performance does not improve monotonically with $K$, and stability shows no consistent association with SSM performance. These findings suggest that intra-method stability is informative but an unreliable indicator of downstream segmentation performance, highlighting the need to evaluate band-selection methods across repeated samples, subset sizes, and SSMs.