Vessel Detection in Synthetic Aperture Radar(SAR) Images Using Deep-Learning

Deep learning models are trained on SAR-Ship-Dataset to compare the detection performance between traditional vessel detection methods e.g. CFAR on SAR images and deep learning detection models.

Dataset

An annotated dataset by SAR experts was recently(2019) published in Remote Sensing journal consisting of 43,819 ship chips is used to evaluate vessel detection "A SAR Dataset of Ship Detection for Deep Learning under Complex Backgrounds" GitHub Paper. This dataset is used to evaluate the detection. We split the dataset into training and evaluation sets. Evaluation set consists of Last 3819 images. Training set consists of first 40000 images.

Models Trained

Faster RCNN

Code for Faster RCNN is cloned from GitHub Repo. To test run demo notebook on subset of images the detection are plotted and saved in results directory.

Retinanet

Code for Retinanet is cloned from GitHub Repo To test run demo notebook on subset of images.