A deep learning approach to ship detection using satellite imagery

Abstract

Abstract Automatic ship detection on remote sensing images is one of the important modules in the maritime surveillance system. Its main task is to detect possible pirate threats as early as possible. Thus, the detection system must be accurate enough as it plays a vital role in national security. Therefore, this paper proposes a deep learning approach to detect the presence of a ship in the harbour areas. DenseNet architecture has been selected as the core convolutional neural network-based classifier, where various finetuning has been done to find the optimal setup. The three hyperparameters that have been fine-tuned are optimizer selection, batch size, and learning rate. The experimental results show a success rate of over 99.75% when Adam optimizer is selected with a learning rate of 0.0001. The test was done on the Kaggle Ships dataset with 4,200 images. This algorithm can be further fine-tuned by considering other types of convolutional neural network architecture to increase detection accuracy.