Developing a Cross-platform Hyperspectral Image Analysis Library with GPU Support

Modern hyperspectral image processing makes use of general purpose Graphical Processing Units (GPUs) to efficiently distribute the workload associated with computing the output of target detection algorithms. Furthermore, state of the art tools that perform operations on hyperspectral sensor data depend on high performance computing power. The core purpose of the project is to develop a cross-platform library, Libdect, which supports target detection algorithms based on hyperspectral image analysis.

Speeding the deployment of hyperspectral image analysis applications, Libdect will replace the time spent developing algorithms that support hyperspectral image processing with a download and configuration of Libdect.

Libdect’s development uses previously implemented detection algorithms, designed for GPUs supported by the Nvidia CUDA architecture. Manipulation of linear algebra equations, corresponding to the Reed-Xiaoli (RX), Matched Filter (MF), and the Adaptive Matched Subspace Detector (AMSD), is essential in order to remove the dependencies inherited from the provided detector implementations. Libdect's detectors are currently supported by free software libraries that perform linear algebra calculations on Linux platforms.

Software Engineering techniques were used to design a portable infrastructure which supports coding guidelines and automated testing of code. At the core of Libdect's infrastructure the CMake development tool provides a cross-platform build system for library configuration. The CTest development tool incorporates automated testing of code, and the KWStyle software was integrated into CMake to test if coding guidelines are being followed.

Libdect is an early prototype of a hyperspectral image analysis library with the potential to complement the development of hyperspectral image applications.