|国家科技期刊平台
首页|期刊导航|计量学报|基于深度学习的墙式氧气吸入器检定方法

基于深度学习的墙式氧气吸入器检定方法OA北大核心CSTPCD

Verification Method of Wall Oxygen Inhaler Based on Deep Learning

中文摘要英文摘要

针对墙式氧气吸入器检定时人工读数影响大、检定效率低等问题,提出一种基于深度学习的墙式氧气吸入器检定方法.通过工业相机拍摄墙式氧气吸入器图像,输入到改进后的残差块ResNet-18网络中,实现对浮子流量计的自动读数.残差块结构改进策略有:在直联通路中增加Dropout网络层;删除残差块中的1×1卷积层;使用LeakyReLU激活函数代替ReLU激活函数.将数据集按照5∶1的比例划分为训练集和测试集,经过100批次的训练,网络模型在测试集上的准确率为98.37%.将中国计量科学研究院检定合格的墙式氧气吸入器连接至检定装置中检定,计算得到浮子流量计的最大示值误差为±0.2L/min,误差在允许范围内,检定结果相同.结果表明,该方法可以代替人工读数,有效提高检定效率.

Aiming at the problems of large influence of manual reading and low verification efficiency of wall oxygen inhaler,a verification method of wall oxygen inhaler based on deep learning is proposed.Images of the wall oxygen inhaler are taken by an industrial camera and fed into the improved residual block ResNet-18 network for automatic readings of the float flow meter.Residual block structure improvement strategies include:add a dropout network layer in the direct connection path;remove the 1 × 1 convolutional layer in the residual block;use the LeakyReLU activation function instead of the ReLU activation function.The dataset is divided into training set and test set according to the ratio of 5∶1,and after 100 batches of training,the accuracy of the network model on the test set is 98.37%.The wall oxygen inhaler with qualified verification results of National Institute of Metrology is connected to the verification device for verification,and the maximum indication error of the float flowmeter is calculated to be±0.2 L/min,the error is within the allowable range,and the verification results are the same.The results show that this method can replace manual reading and effectively improve the verification efficiency.

黄康;孙斌;吴燕娟;裘凯军;赵玉晓

中国计量大学计量测试工程学院,浙江杭州 310018金卡智能集团股份有限公司,浙江杭州 310018浙江省机电产品质量检测所有限公司,浙江杭州 310051

流量计量墙式氧气吸入器检定装置深度学习残差块ResNet-18网络

flow measurementwall oxygen inhalerverification devicedeep learningresidual blockResNet-18 network

《计量学报》 2024 (005)

685-691 / 7

国家自然科学基金(51475440)

10.3969/j.issn.1000-1158.2024.05.11

评论