Extraction and Classification of Flood-Affected Areas Based on MRF and Deep Learning

Floods can cause huge damage to society, the economy, and environment. As a result, it is vital determine extent type of land cover in flooded areas quickly accurately order facilitate disaster relief mitigation efforts. Synthetic aperture radar (SAR) an all-weather, 24 h data source used extract information about flood inundations, its primary aim water body for monitoring. In this study, we have studied backscattering characteristics non-water, combined threshold segmentation method with Markov random fields (MRF), embedded simulated annealing (SA) process image noise reduction, resulting development extraction KI-MRF-SA high accuracy classification automation. Furthermore, object-scale adaptive convolutional neural networks (OSA-CNN) are introduced optical images before provide reference inundation analysis. The proposed study consists following three steps: (1) Kittler Illingworth (KI) thresholding algorithm SAR initial extent; (2) MRF SA algorithms employed as means optimize extent, results across multiple polarizations by using intersection operation final (3) part mapping process, types classified OSA-CNN extents. According experimental results, evident that capable distinguishing from non-water significantly higher (3–5% improvement overall accuracy) than conventional methods. Combined our earlier research, flood-affected could reach 92.7%.