Method for GPU-based spectral data cube reconstruction of integral field snapshot imaging spectrometers

Area of Science:

• Spectroscopy
• Computational Imaging
• Parallel Computing

Background:

• Integral field snapshot imaging spectrometers generate complex spectral data cubes.
• Efficient reconstruction algorithms are crucial for real-time analysis and applications.
• Traditional CPU-based methods present computational bottlenecks for high-throughput data.

Purpose of the Study:

• To enhance the computational efficiency of spectral data cube reconstruction.
• To implement and optimize GPU-based parallel computing for real-time processing.
• To explore applications in agricultural monitoring and crop trait perception.

Main Methods:

• Developed a GPU-accelerated algorithm for spectral data cube reconstruction.
• Applied program parallelization and memory optimization techniques for GPU transfer.
• Compared performance against traditional CPU serial algorithms.

Main Results:

• Achieved an average processing time of 29.43 ms using the GPU algorithm.
• Demonstrated a significant acceleration ratio of approximately 14.27x compared to CPU methods.
• The CPU serial algorithm had an average processing time of 420.46 ms.

Conclusions:

• GPU-based parallelization substantially improves spectral reconstruction efficiency.
• The developed algorithm offers a viable solution for real-time computational demands.
• This technology provides foundational support for advanced agricultural intelligence and monitoring.