YOLO-AES A Lightweight Model for Real-Time Underwater Object Detection

Underwater object detection algorithms are the core of AUV automatic detection and counting. However, the complex underwater environment with dense underwater organisms and small size brings great challenges to the detection task, which is the current research hotspot in the field of computer vision. Based on YOLOv8, this paper proposes A lightweight model for real-time underwater object detection, YOLOv8-AES. The model improves the structure fusion of YOLO model for underwater object detection. Firstly, by introducing the iAFF multi-scale channel attention module in the C2f module, the model has a unified and general feature fusion capability. Then SPPELAN module which assigns different weights to different regions by calculating the correlation between feature maps, so as to realise the local attention enhancement of the model and improve the detection ability of underwater objects. Finally, the Spatial Enhanced Attention Module (SEAM) is introduced to achieve the compensation of the response loss of the occluded part by enhancing the response of the unoccluded part, and to optimise the detection ability of underwater objects with object occlusion. The experimental results show that the proposed lightweight model achieves a mean accuracy (mAP) of $84.7 \%$ on the DUO dataset, and the floating point operations (FLOPs) of the model is only 8.7G, which allows it to achieve a real-time detection speed of 500 frames per second (FPS). This demonstrates that the model can effectively address the challenges of object detection in complex underwater environments.