Multi-GPU Based Parallel Design of the Ant Colony Optimization Algorithm for Endmember Extraction from Hyperspectral...

Area of Science:

• Remote Sensing
• Image Processing
• Computational Intelligence

Background:

• Spectral unmixing is crucial for hyperspectral remote sensing.
• Endmember extraction is a critical step in spectral unmixing.
• Ant Colony Optimization for Endmember Extraction (ACOEE) offers accurate results but suffers from high computational complexity.

Purpose of the Study:

• To enhance the computational performance of ACOEE for hyperspectral data analysis.
• To adapt ACOEE for efficient execution on multi-GPU systems.
• To overcome the limitations of high computational complexity in existing ACOEE algorithms.

Main Methods:

• Proposed a multiple sub-ant-colony-based parallel design for ACOEE.
• Implemented an innovative local pheromone mechanism for sub-ant-colonies.
• Optimized ACOEE for multi-GPU parallel computing to minimize synchronization overhead.

Main Results:

• The proposed parallel ACOEE design significantly improved computational performance.
• Experiments on real hyperspectral datasets validated the effectiveness of the method.
• The multi-GPU implementation achieved substantial speedups without compromising accuracy.

Conclusions:

• The developed parallel ACOEE method effectively addresses the computational challenges of hyperspectral image analysis.
• The local pheromone mechanism facilitates efficient parallel execution on multi-GPU systems.
• This approach enables faster and more scalable endmember extraction for hyperspectral data exploitation.