Publication

Abstract : The identification of ships using satellite imagery has been a subject of continuous research due to its vital importance in maritime surveillance and navigation. It is challenging to recognize ships in satellite photographs due to the wide range of ship sizes and the complex backgrounds. In this study, we propose a system for classifying ships that combines the benefits of two deep-learning models, YOLOv3 and DenseNet. We began using YOLOv3 because it is a well-liked object identification model. However, we discovered that YOLOv3 is unable to recognize small objects, such as miniature ships, in satellite photographs. To get around this problem, we employed DenseNet, which is known for its ability to distinguish both large and small objects. However, DenseNet requires a significant amount of RAM for computations, which might be a disadvantage when resources are few. To get around this limitation, we coupled ResNet with DenseNet, a popular deep learning model recognized for its processing efficiency. As demonstrated by our recommended solution, which effectively recognizes ships in satellite data by combining the advantages of YOLOv3, DenseNet, and ResNet, ResNet and DenseNet working together achieved greater accuracy and more efficient use of memory. Our results demonstrate the efficacy of our approach and the potential applications to real-world issues like maritime surveillance and navigation.