基于改进卷积神经网络的智能物料识别技术研究  

Research on intelligent material identification technology based on improved convolutional neural network

作　　者：杨亚萍 刘军 王洪亮 YANG Yaping;LIU Jun;WANG Hongliang(State Grid Longnan Electric Power Supply Company,Wudu 746000,China;State Grid Gansu Electric Power Company,Lanzhou 730070,China)

机构地区：[1]国网陇南供电公司,甘肃武都746000 [2]国网甘肃省电力公司,甘肃兰州730070

出　　处：《电子设计工程》2024年第8期191-195,共5页Electronic Design Engineering

基　　金：甘肃陇南供电公司2022年管理创新项目(D22FZ2712018)。

摘　　要：针对输电线路中小目标物料识别难度高、准度率低的问题,提出了一种基于改进卷积神经网络的电力线路物料智能识别技术方案。该方案采用改进导向滤波算法和直方图均衡化方法完成了电力线路图像的预处理,对于原始SSD算法存在网络模型复杂、计算速度慢的缺陷,文中采用改进的轻量化卷积神经网络作为基本网络,进一步采用k-means算法优化默认锚框的宽高比,提升了物料识别的准确率。仿真实验结果表明,所提方法相比于原始SSD算法在训练速度和识别准确度方面均具有较大的提升,在实际配电网项目的审计应用中,能够准确识别不同类型的物料,识别精确率大于88%,能够为配电网审计工作提供精准地决策辅助。Aiming at the problems of high difficulty and low accuracy in the identification of small and medium-sized target materials in transmission lines,an intelligent identification technology scheme of power line materials based on improved improved convolutional neural network is proposed in this paper.In this scheme,the improved guided filtering algorithm and histogram equalization method are used to complete the preprocessing of power line image.For the defects of complex network model and slow calculation speed of the original SSD algorithm,this paper uses the improved lightweight convolution neural network as the basic network,and further uses the k-means algorithm to optimize the width height ratio of the default anchor frame,which improves the accuracy of material recognition.Simulation results show that compared with the original SSD algorithm,the proposed method has a great improvement in training speed and recognition accuracy.In the audit application of actual distribution network projects,it can accurately identify different types of materials,and the recognition accuracy is greater than 88%,providing accurate decision-making assistance for distribution network audit.

关 键 词：目标检测 物料审计 卷积神经网络 K-MEANS聚类 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置] TN99[自动化与计算机技术—控制科学与工程]