Making Great Strides Towards Road Detection

A significant amount of research has been carried out in extracting land surface objects, but intelligent digital surface models for monitoring land surface objects are still an active research topic due to emerging technologies such as IoT and Blockchain. These technologies play an important role in quantifying the ecological and geographical properties of the land surface. About such technology of detecting buildings, roads, and terrain from satellite images offer a lot of potential for tracking the migration of large chunks of the population and helps in geographical analysis of the city. In this paper, we explore a Convolutional Neural Network method for extracting land surface objects from satellite imaging with the help of U-Net. As a known fact, the number of disasters occurring every year affects thousands of the population, so suitable mechanisms must be provided for rescue operations. To provide these rescue operations, predictions about the geographical location are of primary importance. Our model produces reasonable accuracy of 60.62% at a very minimal loss rate.

Keywords: U-Net, Spatial Processing, Image Segmentation, Deep Learning, Computer Vision, Down and Up Sampling, Skip Connection.