ROI-Based On-Board Compression for Hyperspectral Remote Sensing Images on GPU.

Area of Science:

• Earth Remote Sensing
• Data Compression
• High-Performance Computing

Background:

• Hyperspectral sensors provide valuable spatial and spectral data for Earth observation.
• Large data volumes from spaceborne sensors necessitate on-board compression to manage storage and transmission limitations.
• On-board data compression faces challenges due to space environment constraints and limited resources.

Purpose of the Study:

• To propose a framework for efficient on-board hyperspectral data compression utilizing GPU parallel computing.
• To implement a target-related compression strategy for optimizing data volume management.
• To enable near real-time analysis and selective compression of regions of interest.

Main Methods:

• Development of a framework leveraging NVIDIA's CUDA architecture for GPU acceleration.
• Implementation of an unsupervised classifier for automatic land cover recognition or event detection.
• Application of space-variant compression techniques including Principal Component Analysis (PCA), wavelet, and arithmetic coding with different bit rates.
• Integration of data volume management for transmission to the Ground Station.

Main Results:

• Demonstration of a GPU-based framework for on-board hyperspectral data compression.
• Successful implementation of unsupervised classification for region identification.
• Effective application of variable bit rate compression tailored to specific regions.
• Validation using real data from an Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) sensor in a harbor area.

Conclusions:

• The proposed GPU-accelerated framework offers an efficient solution for on-board hyperspectral data compression.
• The target-related compression strategy allows for optimized data management and near real-time processing.
• This approach addresses the challenges of limited resources in spaceborne remote sensing applications.