An improved genetic clustering architecture for real-time satellite image segmentation

In the last decade, researchers have focused on the development of hardware architectures for several real-life applications including image segmentation. Accurate analysis of segmented high-resolution satellite image help in identifying flood, fire, cloud, snow, and other natural phenomenon. In this paper, an improved genetic clustering architecture is proposed by introducing innovative architectures for crossover and mutation modules. In this architecture, complexity is low due to the use of Manhattan distance instead of traditional Euclidean distance. Testing of the proposed architecture has been carried out on two satellite captured flood images of Myanmar, Burma 2015, and Chennai, India 2015. Both the satellite images have been successfully segmented and obtained satisfactory PSNR and SSIM values, with an improved power consumption of 31 mW and 191 MHz clock frequency. In comparison with state-of-art architectures, the proposed work delivers satisfactory results in terms of power reduction, clock period, design complexity and resource utilization.