Abstract: To achieve precise detection of underwater biological targets and support the sustainable development of marine resources ,this paper proposes an improved YOLO11n-based detection algorithm. Building upon the YOLO11n baseline model, the algorithm introduces the PPA ( Parallel Patch Attention ) module to enhance feature extraction capability for small underwater targets; employs the Detect_ Efficient module to optimize the detection head and improve multi - scale target detection accuracy; and incorporates the CSFCN feature calibration module to address feature loss caused by the lack of global contextual information during convolution,thereby boosting detection accuracy in blurred underwater images. Compared with the original YOLO11n,the improved model achieves a 1. 9% increase in mAP @ 0. 5 for underwater target detection. When compared to mainstream object detection algorithms, the proposed model also demonstrates superior performance in both precision and recall,reaching 85. 6% and 75. 5% ,respectively. Experiments verify that the improved YOLO11n exhibits better detection performance in underwater target detection tasks compared to mainstream models.

Key words: Underwater biological target detection, deep learning, YOLO11n

摘要： 为实现精准水下生物目标检测 ,保障海洋可持续发展 ,本研究提出一种基于改进 YOLO11n 检测算法 。该算法以YOLO11n 为基础模型 ,通过引入 PPA 并行分块注意力模块 ,增强对水下小目标的特征捕捉能力;采用 Detect_Efficient检测头模块优化检测头 ,提升多尺度目标检测精度 ;借助 CSFCN 特征校准模块解决卷积过程中由于图像全局上下文信息丢失而导致特征丢失的问题 ,提高了模型在图像模糊情况下的检测精度 。与原YOLO11n 对比 ,改进后的模型在水下目标检测时 mAP@ 0. 5 提升了 1. 9 个百分点;改进后的模型与主流目标检测算法相比 ,在精确率和召回率上同样表现更优 ,精确率达 85. 6% ,召回率达 75. 5%。研究表明 ,改进 YOLO11n 与主流目标检测模型相比在水下目标检测任务中有着更好的检测效果。

关键词: 水下生物目标检测, 深度学习, YOLO11n