Intelligent Ship Appearance Detection and Classification Based on Deep Learning

The ship appearance inspection and classification technology can ensure a reduction in maritime accidents and monitor the marine environment, with widespread applications in both civilian and military domains. Technology for intelligent ships has been pushed forward thanks to advancements in deep learning and AI. Consequently, this work investigates deep learning-based algorithms for ship appearance recognition and categorization. In this work, we present a ship recognition system that builds on top of YOLOv5 with several improvements to make it more accurate. Because much of the background in ship photographs is water and sky, the primary network of the algorithm is enhanced with an attention mechanism. This allows for better extraction of ship features, focusing more on these useful features and suppressing other irrelevant ones. Improvements are made to the YOLOv5 network structure, including modifications to its neck structure and enhancements to feature extraction from the backbone to the neck. Additionally, a new fusion module is used in the neck to assign different weights to different feature layers for fusion, achieving better integration of the feature extraction network and enhancing the model's feature extraction and fusion capabilities.