Hurricane and tropical storm inducing rainfall are the main sources of severe urban and riverine floods in South Carolina. In iWERS, we are working on different pieces of a smart early warning system. At the first step, we defined a computer vision project for semantic understanding of visual scenes using Convolutional Neural Networks. In this project, we are developing an extensive water-related object dataset which consists of 5,000 pixel-wise annotated images with 51 different labels. This dataset which is the first exclusive semantic segmented water-related dataset, contains 18 types of natural water system labels, such as river, sea and wetland, and 17 built water-related objects, including dam, culvert, canal, etc. In addition, 21 other objects, such as person, car, road, vegetation, etc. that support contextual reasoning for image recognition are annotated.
Moreover, in order to evaluate riverine flooding in forests, wetlands, riparian areas and to provide flood early warning and monitoring systems using visual sensing and image processing, ASPIRE project, a watershed-scale feasible study, has been proposed. In this project, which includes intensive field survey using terrestrial field sites, field data, images, will be gathered along with correspond streamflow data coming from USGS and Santee Experimental Forest gauges to prepare another natural dataset for training ConvNets to extract streamflow features by image parsing.
