FACO-YOLO: Research on an Improved Underwater Object Detection Algorithm Based on YOLOv11n

Aiming at the problems of complex background, dim light, target overlapping and occlusion, and high equipment deployment requirements in underwater detection, this paper proposes an improved FACO-YOLO underwater object detection algorithm based on YOLOv11n, aiming at improving the detection accuracy while realising the light weight of the network. The residual blocks of the C3K2 module in the backbone network of the model are improved by using convolutional gated linear units and separable convolutions to form the FusionFormer module, which enhances the feature extraction of blurred targets. The ADown module is used to replace the traditional downsampling module in the network, reducing the computational cost while decreasing the loss of feature information. Introduce the CSFCN module to improve the neck network to cope with uneven underwater light and complex background interference. Design the Optidetect detection head, incorporating a lightweight Stem layer and minimalist decoupled heads to optimize computational efficiency and reduce parameter count. The experimental results show that the improved algorithm improves the mAP by 1.5%, reduces the number of parameters by 18.2%, and reduces the GFLOPs by 22% on the RUOD dataset, which verifies the effectiveness of the method.