Yolov8-Lite: An Efficient Lightweight Model for Ship Detection

This paper proposes a lightweight network Yolov8-Lite based on YoLov8. The model introduces a novel C2f-Faster, replacing the original module with Faster-Bottleneck to reduce computational overhead. Additionally, Depthwise Convolutions are incorporated into the backbone network to achieve lightweight design without sacrificing performance. In the neck network, an improved BiFPN-I enhances interaction between shallow and deep features, preserving multi-scale information. Furthermore, a lightweight detection head is designed, considerably reducing the parameters and computational volume. Finally, the Coordinate Attention (CA) is introduced to improve detection accuracy. To validate the optimized model, we conducted experiments on the datasets. The results show that with nearly no loss in accuracy, Yolov8-Lite effectively reduces the number of parameters, GFLOPs, and model size by $74.4 \%, 68.3 \%$, and $71.2 \%$. It outperforms the original and other popular lightweight networks.