Abstract
Synthetic aperture radar (SAR) is one of the most important active imaging systems used in remote sensing. Thanks to SAR and deep learning methods, ship detection can be performed with high performances in recent years. However, using the images of different satellites with changing ship sizes and detecting the ships under complex backgrounds are two challenging tasks that decrease ship detection performance. Since the dimensions of the satellite images are quite high, it is also important to use a fast and lightweight deep learning model. In this paper, we propose the usage of EfficientDet-D0 model to provide a robust and fast solution to the above problems. Experiments were carried out on the Ship-Detection-Dataset that includes nearly 40,000 image patches from Sentinel-1 and Gaofen-3 satellites. EfficientDet-D0 model was compared with Faster R-CNN, RetinaNet, and SSD-MobileNetv2 in terms of 13 different performance metrics, computation times, and visual comparison. The results demonstrate that EfficienDet-D0 model provides the most robust solution to the complex background and multiscale ship size problems.
Keywords
Subject Areas
Citations by Year
Related YTU Activities
Sustainability activities and projects connected to this article.
