Application of UNet Fully Convolutional Neural Network to Impervious Surface Segmentation in Urban Environment from High Resolution Satellite Imagery

Impervious surfaces are traditionally mapped from remotely sensed imagery using image classification algorithms. The surface type is complex in that it consists of many distinct materials, for which image classification and aggregation approaches are generally used to map it. This work explores the use of fully convolutional neural networks (FCNN), specifically, UNet, in mapping these complex features at the pixel level from high resolution satellite imagery. Initial results are promising in both qualitative and quantitative assessment when compared to automated products.