MixerCA: An Efficient and Accurate Model for High-Performance Hyperspectral Image Classification

MixerCA, introduced by researchers Mohammed Q. Alkhatib and Ali Jamali, is a novel lightweight model for hyperspectral image (HSI) classification that combines depthwise convolution, self-attention, and coordinate attention into a unified architecture. Unlike traditional CNNs or vision transformers, MixerCA decouples spatial and channel interactions, maintains consistent resolution throughout the network, and directly processes HSI patches. The model was evaluated on four standard HSI benchmark datasets, demonstrating clear advantages over six competing algorithms, including 2D-CNN, 3D-CNN, Tri-CNN, HybridSN, ViT (Vision Transformer), and Swin Transformer. The paper, accepted for publication in Remote Sensing Applications: Society and Environment, highlights MixerCA's efficiency and accuracy, making it suitable for real-world remote sensing tasks where computational resources are limited.

Hyperspectral imaging captures continuous spectral information across dozens or hundreds of bands, enabling precise identification of terrestrial objects for applications like agriculture, mineral exploration, and environmental monitoring. MixerCA's lightweight design—leveraging depthwise convolutions to reduce parameters and self-attention to capture long-range dependencies—addresses the computational bottlenecks of prior deep learning approaches. By integrating token and channel mixing with coordinate attention, the model achieves high classification accuracy without the heavy compute costs of ViT or Swin Transformer. The open-source code release (available on GitHub) allows researchers and practitioners to replicate results and adapt MixerCA for their own datasets, potentially accelerating deployment in fields like precision farming and disaster response.