An Evolved Cellular Automata Based Approach to Hyperspectral Image Processing

This chapter addresses the problem of processing Hyperspectral images (HI) in real time. It is a relevant problem as working with these types of images usually involves processing very large quantities of data due to their high spectral and spatial resolutions. To achieve real time performance requires resorting to extremely distributed architectures, such as Graphic processing units (GPUs). Most of the algorithms that have been developed for Hyperspectral image (HIS) processing are currently sequential and too cumbersome to achieve the necessary parallelism. A very promising approach for solving this problem is using Cellular Automata (CA) based algorithms as this is an intrinsically distributed computational paradigm that would adapt very snugly to these high performance architectures. The main problem of CAs is that it is necessary to endow them with rule sets which, through iterative cycles of interactions among cells, provide the final desired result. Determining these rule sets is non-trivial. In fact it is a highly difficult task, especially in complex processing operations. Here, the objective is to highlight how this problem can be solved through the use of evolutionary techniques. The proposed algorithm has been named Evolving Cellular Automata (ECA) and it has been tested in two standard operations within hyperspectral imaging: edge detection and segmentation. ECA is applied over synthetic and real HISs and the results are compared to those of well-established techniques, confirming the successful response of this approach and its potential for execution in GPU type architectures. This allows their execution in real time even when the size of the images (both spectrally and spatially) or the speed at which they are taken is large.