基于ACE与YOLOv5 的电力遥感图像检测算法
Electric Power Remote Sensing Image Detection Algorithm Based on Automatic Color Balance and YOLOv5
张弢 1蒋东东 1田喆文 1王艺霖1
作者信息
- 1. 长安大学电子与控制工程学院,陕西 西安 710064
- 折叠
摘要
针对电力遥感图像采集时存在大量浓烟以及数据样本少等问题,提出一种基于对数变换的改进型自动色彩均衡与改进后YOLOv5s模型的遥感图像去雾检测算法,旨在通过提高数据集的图像质量进而提高检测网络的检测精度.构建的改进型自动色彩均衡对电力遥感图像去雾增强,并通过图像质量和特征提取两方面进行了实验数据对比,实验结果表明改进的自动色彩均衡算法优于其它算法.其次,通过YOLOv5s检测算法对增强后的数据集进行训练,引入mosaic数据增强算法,并通过构建ghost卷积模块和NAM注意力模块降低了网络参数、提升了网络检测精度.
Abstract
Aiming at the problems of a large amount of smoke and few data samples in the collection of electric power remote sensing images,an improved automatic color equalization based on logarithmic transformation and an im-proved YOLOv5s model for remote sensing image dehazing detection algorithm are proposed.The image quality in turn improves the detection accuracy of the detection network.Firstly,the constructed improved automatic color equalization enhances the dehazing of electric power remote sensing images,and the experimental data are compared in terms of image quality and feature extraction.The experimental results show that the improved automatic color e-qualization algorithm is better than other algorithms.Secondly,this paper trains the enhanced data set through the YOLOv5s detection algorithm,introduces the mosaic data enhancement algorithm,and reduces the network parameters and improves the network detection accuracy by constructing the ghost convolution module and the NAM attention module.
关键词
电力遥感图像/自动色彩均衡算法/目标检测/多尺度金字塔Key words
Electric power remote sensing image/Automatic color equalization(ACE)algorithm/Target Detection/Multi-scale pyramid引用本文复制引用
基金项目
国家自然科学基金青年项目(61702050)
中央高校基本科研业务费专项重点科研平台水平提升项目(300102321504)
河南省交通运输厅科技项目(2019G-2-5)
陕西省自然科学基础研究计划(2022JM-404)
出版年
2024
NETL
NSTL
万方数据