Water area segmentation of the Yangcheng Lake with SAR data based on improved 2D maximum entropy and genetic algorithm

Synthetic Aperture Radar (SAR) sensors are valuable to flood mapping as radar wavelengths can penetrate cloud cover during floods and are insensitive to daylight. Algorithms that enable an automatic delineation of flooded areas are an essential component of any SAR-based monitoring service. In this paper, efficient image processing method based on two-dimensional (2D) maximum entropy and genetic algorithm (GA) for water area extraction during flood event from the Yangcheng Lake area is proposed. Different from traditional one-dimensional (1D) thresholding method, here the water area mapping takes the information of gray distribution and spatial correlation into account. Considering the higher backscatter coefficient from emergent vegetation in water area, it is not reliable to merely depend on traditional 1D thresholding method with respect to SAR-derived water areas. After introducing Abutaleb's method for image segmentation, GA is improved in evolutionary process of selection, crossover and mutation to evaluate the contribution of microwave data for mapping the extent of water area. Extensive experiments demonstrate that our proposed approach can perfectly improve the anti-noise ability despite of appearance of much occlusion from surrounding vegetation. The method presented here is fairly reliable and leads to a potentially useful data set for direct use in water monitoring.