Abstract: In view of the limitations of deep learning algorithms' generalization and robustness, as well as the high computational cost of full parameter fine-tuning (FPFT) in object detection tasks in complex scenarios, an efficient fine-tuning algorithm for low-rank adaptive parameters based on YOLOv11 (You Only Look Once version 11) was proposed. Firstly, a Low-rank Adaptation (LrAn) module was embedded into the backbone network of YOLOv11. Secondly, three low-rank decomposition algorithms, which include Low-Rank Adaptation (LoRA), weight-Decomposed low-Rank Adaptation (DoRA) and Principal Singular values and Singular vectors Adaptation (PiSSA) were combined. Efficient parameter updates were achieved through weight decomposition and dynamic adjustment mechanisms. Finally, during the training process, most of the pre-trained weights of the YOLOv11 network were kept frozen, and only the low rank matrices generated by the three low rank decomposition algorithms in the LrAn module were trained, reducing the trainable parameter size to 1.56% of the original model. Experiments conducted on the COCO (Common Objects in Context) dataset demonstrate that the proposed algorithm improves the precision, recall and average precision by 4.18, 7.11 and 7.85 percentage points respectively compared with the baseline YOLOv11 algorithm. It can be seen that the proposed algorithm provides an effective technical path for lightweight and efficient fine-tuning of large-scale detection algorithms in resource constrained scenarios.

Key words: YOLOv11, Efficient parameter fine-tuning, Object detection, Low-rank adaptive, Deep learning

摘要： 针对复杂场景下目标检测任务中深度学习算法泛化性、鲁棒性受限以及全参数微调(FPFT)计算成本高的问题，提出一种基于YOLOv11(You Only Look Once version 11)的低秩自适应参数高效微调算法。首先，在YOLOv11骨干和颈部网络中嵌入低秩自适应(LrAn)模块；其次，结合低秩自适应(LoRA)、权重分解低秩自适应(DoRA)和主奇异值与奇异向量自适应(PiSSA)三种低秩分解算法，通过权重分解与动态调整机制实现参数的高效更新；最后，在训练过程中，将YOLOv11网络的绝大部分预训练权重保持冻结状态，仅对LrAn模块中由三种低秩分解算法生成的低秩矩阵进行训练，将可训练参数规模缩减至原算法的1.56%。COCO(Common Objects in Context)数据集实验表明，所提算法相较基线YOLOv11算法在精确度、召回率和平均精度均值指标上分别提升4.18、7.11和7.85个百分点。可见，所提算法为资源受限场景下的大型检测算法轻量化与高效微调提供了有效技术路径。

关键词: YOLOv11, 参数高效微调, 目标检测, 低秩自适应, 深度学习