Deep Learning Approaches for Object Detection in Short-Range GBSAR Images

This paper describes the developed system for detecting objects on low-resolution radar images obtained using Ground-Based Synthetic Aperture Radar (GBSAR). The focus of the work is on the developed methods and algorithms for the object localization, and building and selecting a classifier for use in the system with a limited size dataset and low radar image resolution. In the first step object localization detects and extracts the image parts corresponding to objects and then processes them for use in the classification algorithms. Different classifier algorithms were tested and the results of the algorithm executions are presented and performance metrics on different datasets compared. Based on the results it is shown that even with a very small dataset and a low-resolution system a successful detection can be efficiently achieved.