Navigating the Future: Intelligent Ship Detection through Multisensor Imagery and Deep Learning

Abstract. Satellite ship detection is essential for maritime navigation, illicit activities monitoring, and environmental preservation. Deep learning has greatly increased object detection accuracy, especially for high-resolution satellite and SAR imagery. However, class imbalance, environmental fluctuations, and real-time application require more robust detection methods. The research work aims to increase detection accuracy across environmental conditions and ship types for maritime surveillance and environmental monitoring. Mumbai and Chennai were chosen for their marine qualities for the research. The team tested deep learning models using 1000 high-resolution pictures and SAR data. This involved data preprocessing, annotation, and repeated YOLOv8 and Mask R-CNN model training. YOLOv8, which detects objects in real time, classified ships into 15 categories, while Mask R-CNN segmented. Recall, precision, and mean average precision were assessed. Dataset classes were balanced using RLE. The work segments satellite images for ship recognition using U-Net-based deep learning. Data augmentation and loss functions like binary cross-entropy and Dice coefficient improved detection. The model identified ships under difficult conditions with a Dice Coefficient of 0.86, Precision of 89%, and recall of 82%. However, feeble or partially visible vessels are still difficult to detect. This segmentation-based technique improves maritime surveillance, environmental conservation, and national security, requiring more research and execution. Model performance was measured using important metrics. Mask R-CNN, with a ResNet50 backbone and a learning rate of 2.75e-05 to 2.75e-04, had an Average Precision Score of 22.08% for Mumbai but struggled in real-time applications. YOLOv8 outperformed with 52.8% Precision, 41.9% Recall, 44.2% mAP@50, and 26.3% mAP@50-95. Integration of SAR data enhanced detection accuracy in various environments. Oil tankers and cruise ships were precise, but rafts and dinghies were harder to spot. This shows that YOLOv8 can detect ships in real time, making it a feasible tool for port surveillance and marine traffic monitoring. The hybrid technique, which combines YOLOv8's speed with Mask R-CNN's segmentation, may recognize smaller ship classes better. To overcome maritime object detection issues, future research should extend datasets, refine model architectures, and use sophisticated deep learning approaches.