多尺度特征融合增强检测模型MFFE-YOLO
A multi-scale feature fusion enhanced detection model MFFE-YOLO
彭继慎 1马龙泽 1孙梦宇 1刘金龙2
作者信息
- 1. 辽宁工程技术大学 电气与控制工程学院,辽宁 葫芦岛 125105
- 2. 辽宁工程技术大学 软件学院,辽宁 葫芦岛 125105
- 折叠
摘要
为解决传统巡检图像检测方法对电力设备小目标缺陷检测能力弱、错检和漏检率高、浅层网络语义信息不足等问题,提出针对电力设备小目标缺陷的多尺度特征融合增强检测模型(multi-scale feature fusion enhanced you only look once,MFFE-YOLO).该方法设计了一种多尺度特征融合增强机制(multi-scale feature fusion enhancement,MFFE),能够更全面地捕捉目标特征.研究表明:在C2f-EF模块中嵌入跨空间学习多尺度注意力机制EMA以及FasterNet Block,能够优化模型的运行效率;MFFE-YOLO方法的平均精度、参数量和帧率指标均优于其他方法,能够实现高精度与实时性之间的良好平衡.
Abstract
In order to solve the problems of weak detection ability of small target defects of power equipment,high false detection and missed detection rate,and insufficient shallow network semantic information in traditional inspection image detection methods,a feature fusion enhanced detection method for small target defects of power equipment(multi-scale feature fusion enhanced you only look once,MFFE-YOLO)is proposed.This method designs a multi-scale feature fusion enhancement mechanism(multi-scale feature fusion enhancement,MFFE),which can capture target features more comprehensively.The study shows that embedding the cross-space learning multi-scale attention mechanism EMA and FasterNet Block in the C2f-EF module can optimize the operation efficiency of the model;the average accuracy,parameter amount and frame rate indicators of the MFFE-YOLO method are better than that of other methods,and can achieve a good balance between high accuracy and real-time performance.
关键词
电力巡检/电力设备缺陷/小目标检测/特征融合增强/YOLO/多尺度特征Key words
power inspection/power equipment defects/small target detection/feature fusion enhancement/YOLO/multiscale feature引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据