BCG-Former: Toward Pareto-Efficient Hyperspectral Image Classification via Band-Contextual Gating
Quick Answer
BCG-Former is a lightweight CNN-Transformer hybrid for hyperspectral image classification, achieving up to 99.49% accuracy with sub-millisecond latency and minimal parameters.
Quick Take
It excels in balancing accuracy and efficiency, outperforming traditional methods while maintaining a computational cost advantage.
Key Points
- BCG-Former achieves 91.51%-99.49% accuracy on multiple benchmark datasets.
- It maintains inference latency between 0.91-0.95ms with only 0.10-0.23M parameters.
- The model introduces Band-Contextual Gating for adaptive spectral recalibration.
- Ablation studies confirm the complementary nature of its three innovative components.
- BCG-Former is positioned on the Pareto frontier of accuracy versus latency.
DeepSignal Analysis
What happened
BCG-Former is a newly proposed lightweight CNN-Transformer hybrid model for hyperspectral image classification. It achieves high accuracy, reaching up to 99.49%, while maintaining low latency and minimal parameters, making it suitable for resource-constrained environments.
Key evidence
- BCG-Former achieves overall accuracy ranging from 91.51% on the Houston 2018 dataset to 99.49% on the Houston 2013 dataset.
- The model maintains sub-millisecond inference latency, specifically between 0.91 and 0.95 milliseconds, while using only 0.10 to 0.23 million parameters.
- Ablation studies indicate that all three innovations in BCG-Former are complementary, with Band-Contextual Gating contributing the most to its performance.
Why it matters
The development of BCG-Former addresses the growing need for efficient hyperspectral image classification systems that can operate under strict computational constraints, such as those found in UAVs and small sensors. Its ability to balance accuracy and efficiency could enhance real-time remote sensing applications significantly.
What to watch
Paper Resources
Source Excerpt
Hyperspectral image (HSI) classification systems are increasingly deployed on platforms with strict computational budgets, such as UAVs and small spaceborne sensors. In these settings, accuracy alone is not enough; the model must also run within tight latency and memory constraints. Most recent HSI classifiers, however, focus on accuracy and pay relatively little attention to these constraints. We propose BCG-Former, a lightweight CNN-Transformer hybrid that targets this trade-off. The model int
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.