U-net based MRA framework for segmentation of remotely sensed images

Segmentation of remotely sensed images lies in the intersection of the domains of remote sensing and computer vision. It is used to systematically extract information from data collected by various airborne and spaceborne sensors, resulting in a simpler representation. This method is used in various applications which include change detection, land cover and land use classification, resource exploration, the study of natural hazards, and mapping. This paper proposes a composite architecture for segmentation of remotely sensed images derived from the U-Net aided by processing the information obtained using multi-resolution analysis (MRA). This increases the contextual overview of the network as the same data on multiple scales is available for feature extraction and learning. This work further analyzes the effects of different wavelets for decomposition on segmentation performance. On comparison with the plain U-Net model performance, it is observed that the MRA aided network can segment areas with higher accuracy and mean intersection over union, the best performance being with Daub basis of MRA.