Vessel Detection Based on Ocean Satellite Data Using U-Net

Abstract Effective maritime traffic management requires technologies capable of detecting and monitoring vessels navigating vast ocean areas. Optical sensor images from ocean satellites provide high-resolution data for intuitive interpretation. When combined with Convolutional Neural Networks (CNN), one of the most significant architectures in deep learning, they can achieve excellent object recognition performance. This study proposes a detection method using optical sensor images based on the U-Net model. U-Net is a fully convolutional network-based model proposed for image segmentation. U-Net can apply various backbones to the encoder-decoder path to enhance detection performance, and the detection performance results were compared using VGG Net, ResNet, DenseNet, and EfficientNet. This study is expected to serve as a methodology for future vessel detection using ocean satellite optical sensors.