FPGA mapping for data processing of hyperspectral images

Hyperspectral imaging is used for surveillance and other real-time applications. Hyperspectral imaging applications typically require high computational accuracy with relatively high-power consumption. Hyperspectral image processing presently exists on a GPU, but has drawbacks like memory access overlap, memory optimization, high power consumption and additional microprocessor requirement. Efficient FPGA mapping for hyperspectral image data processing uses parallel processing techniques to overcome the drawbacks of GPU and simultaneously increase computational speed. In this work computations are performed using various MATLAB functions such as averaging, covariance, eigenface, and others. PCA and Euclidean distance algorithms are implemented for hardware design and implementation. Computations result in matching the test images with one of the train set images. The designed FPGA mapping for hyperspectral image data processing is executed and implemented on Xilinx Zynq Field Programmable Gate Array (FPGA) with VHDL coding for testing and uses MATLAB for database creation. The FPGA mapping utilizes the CPU for particular processes, and a SOC for specific processes as required. The FPGA mapping has the advantage of utilizing the system main CPU and its cache memory for hyperspectral image processing to enable the algorithm to gain performance time improvements in real time execution. All steps of processing for image recognition are presented in detail