Research on Fish School Quantity Detection Algorithm Based on HyperC2Net+MANet Improved YOLO11

This paper focuses on solving the problem of insufficient detection accuracy in fish school detection caused by dense fish schools and complex underwater environments. A fish school detection algorithm based on HyperC2Net+MANet improved YOLO11 is proposed. HyperC2Net is developed using the HGC-SCS framework as the neck network for object detection. It can achieve high-order message passing between the semantic layer and positions, operate at five scales, break the traditional grid structure, and promote complex high-order interactions between different layers and positions. This significantly improves the ability of the neck to extract high-order features, enabling the model to better capture the features of fish schools. At the same time, the Mixed Aggregation Network (MANet) is introduced into the backbone network. This network contains three types of modules, which enriches the information flow and enhances the feature extraction ability of the backbone network, helping the model to focus more accurately on the key information of fish schools. Experiments were carried out on a self-built fish school detection dataset. The results show that the improved algorithm achieves an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP} {@} 0.5$</tex> of 96.6 % and an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP} {@} 0.5$</tex>: 0.95 of 72.4 %. Compared with the original YOLO11, the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP} {@} 0.5$</tex> of the improved model is increased by 5.81 %, and the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP} {@} 0.5: 0.95$</tex> is increased by 17.72 %. In addition, the algorithm's inference speed reaches 53.2FPS, meeting the application requirements for real-time fish school detection, and has good application prospects and potential in underwater fish school detection scenarios.