基于改进YOLOv8n的数字式仪表读数识别

Digital Meter Reading Recognition Based on Improved YOLOv8n

摘要:

针对数字式仪表读数中人工抄表误差大和识别准确率低的问题，提出一种基于改进 YOLOv8n 的数字式仪表自动识别算法。在网络结构中，将 SCConv 模块嵌入主干网络的 C2f 结构，以增强模型在空间与通道维度的特征提取能力，提高对细粒度目标和边缘细节的感知效果；使用 Focal Modulation 结构替换原有的 SPPF 模块，通过分层上下文化与门控聚合机制实现局部与全局特征的高效融合，从而提升检测精度与特征表达能力；在损失函数设计方面，将 CIoU 损失替换为 EIoU 损失，以优化边界框回归的收敛速度与定位精度。实验结果表明，该方法在自建仪表数据集上的 mAP@0.5 较原始 YOLOv8n 模型提升 1.5%，其中小数点类别精度提升 14.3%。结果验证了所提算法在数字式仪表读数任务中的有效性与优越性。

Abstract:

To address the problems of large manual reading errors and low recognition accuracy in digital meter reading, this paper proposes an improved YOLOv8n-based automatic recognition algorithm for digital instruments. In the network structure, the SCConv module is embedded into the C2f blocks of the backbone to enhance the feature extraction capability in both spatial and channel dimensions, thereby improving the perception of fine-grained targets and edge details. Meanwhile, the original SPPF module is replaced with the Focal Modulation structure, which achieves efficient fusion of local and global features through hierarchical contextualization and gated aggregation mechanisms, thus enhancing detection accuracy and feature representation capability. In terms of the loss function, the CIoU loss is replaced by EIoU loss to accelerate the convergence of bounding box regression and improve localization precision. Experimental results show that the proposed method improves the mAP@0.5 by 1.5% compared with the original YOLOv8n model on a self-built meter dataset, with the accuracy of the decimal point class increasing by 14.3%. These results demonstrate the effectiveness and superiority of the proposed algorithm in digital meter reading tasks.