摘要
为实现电力巡检过程的智能化,解决拍摄的巡检照片质量不高、分类准确率低的问题,本研究提出一种基于改进YOLOv3的电力巡检照片分类命名方法。该方法利用融合通道洗牌和通道注意力提高模型的特征表征能力,结合多层感知器神经网络与后处理模块完成最终的分类命名。实验结果表明,改进后的YOLOv3模型平均准确率优于原YOLOv3模型,平均精度均值由85.62%提高至94.73%;与现有的主流分类模型相比,能更好地处理拍摄质量差的巡检照片,提升电力巡检效率。
To achieve intelligent power inspection process and solve the problem of poor quality and low classification accuracy of inspection photos taken,a classification and naming method for inspection photos based on improved YOLOv3 was proposed.The modules of fused channel shuffling and channel attention were used to improve the model’s feature characterization ability.Then,the final classification and naming task was completed by combining multiplayer perceptron neural network and post-processing module with the model.The experimental results show that the improved YOLOv3 model has better average accuracy than the original YOLOv3 algorithm,with the mean average precision YOLOv3 increasing from 85.62%to 94.73%.Compared to mainstream image classification models,the improved model can better handle poorly captured inspection photos,thus improving the efficiency of the power inspection process.
作者
郑高
郑恩辉
王桂荣
ZHENG Gao;ZHENG Enhui;WANG Guirong(College of Mechanical and Electrical Engineering,China Jiliang University,Hangzhou 310018,China)
出处
《山东科技大学学报(自然科学版)》
北大核心
2025年第3期107-118,共12页
Journal of Shandong University of Science and Technology(Natural Science)
基金
国家自然科学基金项目(61203113)
浙江省自然科学基金项目(LGG22F030001)。
关键词
电力巡检
深度学习
YOLOv3
目标检测
注意力机制
electric power inspection
deep learning
YOLOv3
object detection
attention mechanism
