An FPGA-oriented HW/SW Codesign of Lucy-Richardson Deconvolution Algorithm for Hyperspectral Images
Karine Avagian, Milica Orlandić, Tor Arne Johansen · 2019
Hyperspectral images acquired during remote sensing are usually corrupted by optical blur and additive noise. This article considers a method to reduce the spatial blur in hyperspectral images by applying the Lucy-Richardson deconvolution algorithm. The algorithm is based on convolution with a point-spread function, which is assumed to be known. The complete algorithm is implemented as a HW/Sw Co-Design between the central processing unit and the programmable logic. The convolution, as a time consuming operation, is accelerated in an FPGA. When compared to the software implementation, a speed-up by a factor of 2.7 is achieved when using the accelerated HW/SW Co-Design implementation. Hyperspectral images synthetically distorted by a known Gaussian blur and an additive noise are used for testing. The images are then restored by applying the Lucy-Richardson deconvolution algorithm on each spectral band.