Asymmetric coordinate attention spectral-spatial feature fusion network for hyperspectral image classification

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Deep learning hyperspectral classification algorithms are gaining attention.
• Existing models suffer from high complexity and time consumption.
• There's a need for improved accuracy and reduced complexity in hyperspectral image classification.

Purpose of the Study:

• To propose an efficient and accurate hyperspectral classification network.
• To reduce model complexity and computational time.
• To capture distinguishing spectral-spatial features for improved classification.

Main Methods:

• Developed an asymmetric coordinate attention spectral-spatial feature fusion network (ACAS2F2N).
• Introduced adaptive asymmetric iterative attention for discriminative feature extraction.
• Integrated coordinate attention and a strip pooling module for enhanced feature representation.

Main Results:

• The ACAS2F2N achieved state-of-the-art performance on IP, KSC, and Botswana datasets.
• Demonstrated lower time complexity compared to existing methods.
• The proposed feature fusion method is adaptable to various skip connection tasks without manual parameter tuning.

Conclusions:

• The ACAS2F2N effectively enhances hyperspectral image classification accuracy.
• The network offers a computationally efficient solution for hyperspectral data analysis.
• The novel attention and feature fusion mechanisms contribute to superior performance.