Maritime Ship Target Detection Based on the YOLOv7 Model

Abstract

Maritime ship target detection plays a significant role in the modern maritime domain. However, challenges such as low detection accuracy, false positives, and missed detections hinder its effectiveness due to the complexity of the maritime environment and the diversity and variability of ships. To address this challenge, we propose a sea ship target detection method based on YOLOv7. Firstly, we replace the main neural network with MobileNetv1, reducing model parameters and enhancing inference speed. Secondly, in the feature fusion section, we introduce a shallow feature fusion pathway to comprehensively capture image details and contextual information, improving the detection capability for objects of various scales. Experimental data demonstrates that the enhanced YOLOv7 model has achieved significant results, achieving an average precision of 97.16%, which represents a 5.53% improvement over the initial model. This validates the feasibility of our approach.