A Convolutional Neural Network Model for Superresolution Enhancement of UAV Images

In recent years, the use of unmanned air vehicles (UAVs) in various fields has become widespread. These UAVs have a set of sensors that allow obtaining information of the scenarios by which they fly, in order to monitor them or to be used in their navigation tasks. The camera is one of the most relevant elements. UAVs have limitations regarding their autonomy time. If you want to monitor a large geographical area in a short time, you need to make flights at a higher altitude. This implies loss of spatial resolution, since the cameras themselves have their limitations in terms of their optics and the size of the pixels. In recent years, superresolution techniques based on deep Convolutional Neural Network (CNN) have been developed, which are able to learn the correspondence between low resolution images with their high-resolution counterparts. The problem of these models lies in the computation requirements that they need for their execution, not being viable to be executed in the embedded hardware of a UAV. In this work we propose a superresolution method based on deep CNNs capable of being executed in the on-board equipment of an UAV.