Road Detection and Segmentation from Aerial Images Using a CNN Based System

This paper proposes a system architecture based on deep convolutional neural network (CNN) for road detection and segmentation from aerial images. These images are acquired by an unmanned aerial vehicle implemented by the authors. The algorithm for image segmentation has two phases: the learning phase and the operating phase. The input aerial images are decomposed in their color components, preprocessed in Matlab on Hue channel and next partitioned in small boxes of dimension 33 × 33 pixels using a sliding box algorithm. These boxes are considered as inputs into a deep CNN. The CNN was designed using MatConvNet and has the following structure: four convolutional layers, four pooling layers, one ReLu layer, one full connected layer, and a Softmax layer. The whole network was trained using a number of 2,000 boxes. The CNN was implemented using programming in MATLAB on GPU and the results are promising. The proposed system has the advantage of processing speed and simplicity.