Detection Process of Ships in Aerial Imagery Using Two Convnets

In this article, we focus on boat real time detection in high resolution aerial images taken by UAVs, using convolutional neural networks. This problem is similar to object detection in images taken from the terrestrial surface at ground level or on elevations, but present certain specificities regarding: the smaller size of objects to be detected and camera orientations that are different. As a result, the methods designed for the general detection purpose are not adapted to our problem. Some of these convolutional network-based methods have been adapted for locating objects on remote sensing images by making changes and enhancements on the networks' models and parameters. In this article, we will take a look at some of these techniques and architectures. We will also draw up a comparison table between the various improvements used in these methods. We will also present the strategy and techniques we will adopt in order to provide practical advantages and a reliable solution to our problem. And finally, we will expose the results obtained using our approach.