Automatic Ship Detection and Classification from Imagery Dataset MASATI Using Deep Learning Framework

A significant challenge is the automatic detection and classification of ships from aerial images for maritime safety and surveillance. There are two techniques, image processing and computer vision, used before the traditional supervised classifiers to extract the features from the images and classify them. The framework proposed in this paper is You Only Look Once (YOLOv8), which was recently developed and implemented for object detection and classification on the COCO dataset. The authors used the framework with some preprocessing and augmentation on the satellite image to detect and classify ships, which helps maritime organizations locate the objects. The MAritime SATellite Imagery dataset (MASATI) of more than 7000 images is used in this paper. The proposed method is tested with two different preprocessing steps, auto-orient and mosaic augmentation, in version 1 and version 2; the preprocessing step includes auto-orient and grayscale along with the augmentation as crop, shear, blur, and mosaic. The framework generates mAP of 57.8% and 63.7% for version 1 and version 2, respectively.