一种改进YOLOv11n的电表作业缺陷检测模型

An improved YOLOv11n model for detecting defects in electricity meter operations

摘要:

针对传统电表作业缺陷检测方法检测效率低、检测准确率不足以及深度学习算法对计算资源要求高的问题,提出一种基于改进YOLOv11n的轻量化检测模型。该模型通过使用MobileNetV4替换原主干网络来降低计算冗余,大幅降低模型参数量和计算量；设计LBCFF特征融合网络,在轻量化模型的同时增强模型多尺度特征交互,提升检测精度；并通过引入CGA注意力机制优化C3k2模块,提升模型在复杂背景下的目标区分能力。实验结果表明,改进后模型的mAP达98.1%,较基准模型参数量减少了60.85%、计算量降低47.62%、模型体积压缩51.20%,同时mAP提升1%,并且满足实时检测的要求,证明该模型在保持高精度检测的前提下显著降低资源需求。

Abstract:

To address the limitations of low detection efficiency, insufficient accuracy, and high computational resource demands inherent in traditional electricity meter defect detection methods, this paper proposes a lightweight detection model based on an improved YOLOv11n architecture. This model reduces computational redundancy by replacing the original backbone network with MobileNetV4, significantly reducing the number of model parameters and computational load; a LBCC feature fusion network is designed to enhance multi-scale feature interaction while lightweighting the model, thereby improving detection accuracy; and the CGA attention mechanism is introduced to optimize the C3k2 module, enhancing the model"s ability to distinguish targets in complex backgrounds. Experimental results demonstrate that the improved model achieves an mAP of 98.1%. Compared to the original model, it reduces the number of parameters by 60.85%, computational complexity (FLOPs) by 47.62%, and model size by 51.20%, while simultaneously improving mAP by 1% point. Furthermore, the model meets real-time detection requirements. These findings confirm that the proposed approach significantly reduces computational resource demands while maintaining high detection accuracy..