基于改进的YOLOX家用燃气表检测算法  

Digital detection algorithm of household gas meter based on improved YOLOX

作　　者：李天顺 孙佳东 李东洋 姜衍超 朱建军[1] LI Tianshun;SUN Jiadong;LI Dongyang;JIANG Yanchao;ZHU Jianjun(School of information and control engineering,Jilin Institute of chemical technology,Jilin 132022;Hangzhou Dechuang Energy Equipment Co.,Ltd.Hangzhou 311100)

机构地区：[1]吉林化工学院信息与控制工程学院,吉林吉林132022 [2]杭州德创能源设备有限公司,浙江杭州311100

出　　处：《长江信息通信》2022年第6期142-146,共5页Changjiang Information & Communications

摘　　要：针对家用燃气表人工集中抄表作业过程中检测背景复杂、实时性差、数字目标小等问题,提出了一种基于改进的YOLOX家用燃气表数字检测算法S-YOLOX。首先,削减CSPDarknet结构,降低主干网络的深度;其次,对特征金字塔进行调整,加强小目标物体检测;最后,修订PANet和Head结构的激活函数,解决梯度轻微爆炸问题。实验结果表明,SYOLOX算法不仅减少了模型的参数量,而且提高了检测精度,其中S-YOLOX-l模型平均精度(mAP)达到了94.9%,提高了2.2%,模型大小仅为原模型的32.9%。Aiming at the problems of complex detection background, poor real-time performance and small digital target in the process of manual centralized reading of household gas meter, a digital detection algorithm S-YOLOX based on improved YOLOX household gas meter is proposed. Firstly, reduce the structure of CSPDarknet and reduce the depth of backbone network;Secondly,the feature pyramid is adjusted to strengthen the detection of small target objects;Finally, the activation functions of PANet and Head structures are revised to solve the problem of gradient slight explosion. The experimental results show that S-YOLOX algorithm not only reduces the number of parameters of the model, but also improves the detection accuracy. Among them, the average accuracy(mAP) of S-YOLOX-l model is 94.9%, increased by 2.2%, and the size of the model is only 32.9% of the original model.

关 键 词：家用燃气表 目标检测 YOLOX CSPDarknet 特征金字塔 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]