EN-YOLO: A Ship Enhanced Detection Network Based on YOLOv8

Efficient mobile monitoring devices are crucial for maritime safety, but deep-learning based ship recognition models are limited in their application on mobile devices due to low detection accuracy and high computational requirements. Therefore, we offer a strengthened detection network based on YOLOv8, named EN-YOLO. Firstly, we introduced efficient multi-scale attention (EMA) into the backbone network to achieve cross channel information crossover and cross spatial learning to strengthen the ability to extract features from the backbone network. Secondly, we used a lightweight and efficient SlimNeck module in the neck network to lessen the complexity of the calculation and parameter count of the model while ensuring its accuracy remains the same or improves. This study was validated on the publicly accessible dataset SeaShips, and the trials outcomes showed that the proposed EN-YOLO implemented on the basis of the original model [email protected] boosted by 0.3%, [email protected] 0.95 enhance by 0.7% and reduced computational complexity by 0.2%.