LS-YOLO: A Light YOLO Model for Coastal Ship Target Detection

High-precision target detection of coastal ships is a crucial auxiliary means of safeguarding maritime transportation and enhancing navigation safety. The identification and detection of coastal ships face challenges due to complex sea surface backgrounds, variations in ship scale, and inter-ship occlusion, which result in lower detection accuracy. Additionally, most target detection models based on deep learning suffer from a large number of parameters, making deployment difficult. To address these two issues, this paper proposes a lightweight ship target detection model called LS-YOLO, based on YOLO (You Only Look Once) v8. To reduce the weight of the model's parameters, the conventional convolutional layer in the backbone is replaced with the lightweight GhostConv, and the standard Bottleneck is substituted with the GhostBottleneck. Furthermore, to enhance the important feature information in images, the Convolutional Block Attention Module (CBAM) attention mechanism is incorporated before the large, medium, and small detection heads. We also replace the original CIoU Loss with the Focal EIoU Loss, which improves the model's detection and localization accuracy. Simulation experiments were conducted using the Seaships ship image dataset. The experimental results indicate that, compared to YOLOv8n, the LS-YOLO model improves detection accuracy by 5.7% and recall rate by 2.7%. Simultaneously, the mean Average Precision at 0.5 (mAP0.5) and the mean Average Precision thresholds of 0.5 to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.95(\text{mAP} 0.5: 0.95)$</tex> are enhanced by 3.9% and 7.5%, respectively. Moreover, the number of model parameters is reduced by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{4 5. 3 \%}$</tex>. These results demonstrate that LS-YOLO can maintain high detection accuracy while reducing the complexity of the ship target detection model.