Shanghai Power Station Auxiliary Equipment Works (referred to as SDF) was established in April, 1980 on the base of the urban area previously owned by Shanghai Boiler Works so as to facilitate
1 Project overview The Shasan station of Phase II of Shenzhen’s urban rail transit Line 12 is situated in Bao’an District,Shenzhen.It comprises a two-level underground island platform station,measuring 212 m in leng...1 Project overview The Shasan station of Phase II of Shenzhen’s urban rail transit Line 12 is situated in Bao’an District,Shenzhen.It comprises a two-level underground island platform station,measuring 212 m in length,and 22.6 m in width,with an overburden thickness of about 7.0 m.Fig.1 illustrates the presence of a large underground reinforced concrete stormwater culvert,measuring 11.5 m by 3.6 m,traversing the station’s center.展开更多
针对换流站多种电气设备检测时背景复杂干扰性强而又需要快速准确检测出故障的实际情况,提出基于改进YOLOv5(You Only Look Once)的检测方法。首先,为提高算法的准确性和收敛速度,通过K-means聚类算法对YOLOv5模型中的锚框预设进行改进...针对换流站多种电气设备检测时背景复杂干扰性强而又需要快速准确检测出故障的实际情况,提出基于改进YOLOv5(You Only Look Once)的检测方法。首先,为提高算法的准确性和收敛速度,通过K-means聚类算法对YOLOv5模型中的锚框预设进行改进,在数据集预处理阶段得到更适用于换流站电气设备的锚框,使其更加契合换流站电力设备数据集;然后,为提高算法检测过程的识别速度,在特征提取网络添加注意力机制模块,筛选出重要的特征信息。将改进后的算法网络识别效果与YOLOv5中的原始算法网络检测结果进行对比分析。结果表明,检测平均识别精度均值由71.16%提高至92.51%,检测速度由21帧/s提升至31帧/s;同时与R-CNN(Regions with convolutional neural networks)等算法相比,检测精度与速度都有较大提升。添加可解释性分析,将识别结果通过热力图的形式显示,可以更好地应对算法的潜在风险。展开更多
文摘Shanghai Power Station Auxiliary Equipment Works (referred to as SDF) was established in April, 1980 on the base of the urban area previously owned by Shanghai Boiler Works so as to facilitate
基金This engineering is a demonstration project for Key Research and Development Project of Guangdong Province under Grant No.2019B111105001part of research related to this engineering was financially supported by the project.
文摘1 Project overview The Shasan station of Phase II of Shenzhen’s urban rail transit Line 12 is situated in Bao’an District,Shenzhen.It comprises a two-level underground island platform station,measuring 212 m in length,and 22.6 m in width,with an overburden thickness of about 7.0 m.Fig.1 illustrates the presence of a large underground reinforced concrete stormwater culvert,measuring 11.5 m by 3.6 m,traversing the station’s center.
文摘针对换流站多种电气设备检测时背景复杂干扰性强而又需要快速准确检测出故障的实际情况,提出基于改进YOLOv5(You Only Look Once)的检测方法。首先,为提高算法的准确性和收敛速度,通过K-means聚类算法对YOLOv5模型中的锚框预设进行改进,在数据集预处理阶段得到更适用于换流站电气设备的锚框,使其更加契合换流站电力设备数据集;然后,为提高算法检测过程的识别速度,在特征提取网络添加注意力机制模块,筛选出重要的特征信息。将改进后的算法网络识别效果与YOLOv5中的原始算法网络检测结果进行对比分析。结果表明,检测平均识别精度均值由71.16%提高至92.51%,检测速度由21帧/s提升至31帧/s;同时与R-CNN(Regions with convolutional neural networks)等算法相比,检测精度与速度都有较大提升。添加可解释性分析,将识别结果通过热力图的形式显示,可以更好地应对算法的潜在风险。
