SAR Image Water Area Segmentation Based on Deep Learning

Abstract

SAR remote sensing water segmentation technology can play an essential role in many scientific research and engineering fields. Feature selection and construction are crucial in-ground feature discrimination and classification, affecting the water region's accuracy and effect and land classification. In order to improve the regional consistency of the classification results, this paper proposed a Markov Random Field (MRF) model based on deep learning and Wishart distance and combined with a convolutional neural network to achieve SAR image terrain classification. The results show that the proposed algorithm has advantages. It improves the accuracy and improves the global equilibrium of water extraction. The algorithm has been proved to be effective in monitoring water areas in the lower reaches of the Yangtze River.