摘要
为实现绝缘子及其缺陷实时检测,文章以改进的区域卷积神经网络(Faster Region Convolutional Neural Networks,Faster R-CNN)作为基础研究算法,将残差网络和特征金字塔网络相融合作为特征提取网络,使用深度可分离卷积替换原有的普通卷积,引入SE通道注意力模块,同时改进了网络中的激活函数。改进后的Faster R-CNN与普通Faster R-CNN相比,全类别平均正确率(mean Average Precision,mAP)和检测速度都有所提高。
In order to achieve real-time detection of insulators and their defects,the paper uses the improved Faster Region Convolutional Neural Networks(Faster R-CNN)as the basic research algorithm,fuses the residual network and the feature pyramid network as the feature extraction network,uses the depth separable convolution to replace the original The Faster R-CNN is a basic research algorithm that combines the residual network and the feature pyramid network as the feature extraction network,replaces the original ordinary convolution with the deep separable convolution,introduces the SE channel attention module,and improves the activation function in the network.The improved Faster R-CNN has improved the mean Average Precision(mAP)and detection speed compared with the normal Faster R-CNN.
作者
刘先梁
徐建
郝沛健
刘芷馨
LIU Xianiang;XU Jian;HAO Peijian;LIU Zhixin(School of Intelligent Science and Engineering,Hubei University for Nationalities,Enshi Hubei 445000,China)
出处
《信息与电脑》
2023年第6期78-81,共4页
Information & Computer
关键词
残差网络
特征融合
深度卷积
通道注意力
residual network
feature fusion
depth wise convolution
channel attention