基于注意力机制与多尺度融合的绝缘子缺陷检测
Insulator Defect Detection Based on Attention Mechanism and Multi-scale Fusion
詹志翔1
作者信息
- 1. 温州大学电气与电子工程学院,浙江 温州 325035;温州大学智能锁具研究院,浙江 温州 325036
- 折叠
摘要
针对绝缘子缺陷尺度小且易受到背景干扰较多,检测效果并不理想的问题,提出一种基于YOLOv8n轻量模型的绝缘子缺陷检测方法,该方法结合了注意力机制和多尺度融合技术,达到提升精度与减少模型参数量的效果.首先,在特征提取网络中添加注意力模块.其次,使用添加了小目标检测层的加权双向特征金字塔网络(BiFPN).通过实验表明,所提方法相较于原有方法,在减少模型参数量的同时使平均检测精度提升了 1.9%,验证了该方法的有效性.
Abstract
Aiming at addressing the suboptimal detection performance for insulator defect due to small scaleand suscepti-bility to background interference,the present work proposed a light-weight method of detecting insulator defects based on the YOLOv8n model.This method combines attention mechanism and multi-scale fusion technology to achieve improved accuracy and reduced model parameters.First an attention module was integrated into the feature extraction network.Sec-ond a weighted bi-directional feature pyramid network(BiFPN)with an added small target detection layer was employed.The proposed method was demonstrated by experiment to achieve a 1.9％increase in average detection accuracy while re-ducing model parameters,compared to the original approach,highlighting its effectiveness.
关键词
绝缘子/缺陷检测/YOLOv8n/BiFPN/注意力机制Key words
insulator/defect detection/YOLOv8n/BiFPN/attention mechanism引用本文复制引用
出版年
2024
NETL
NSTL
万方数据