摘要
针对目前工业上目标检测任务较多,却限于设备原因无法流畅运行常规大型目标检测网络,对轻量型目标检测网络需求较大的问题,提出一种新型的轻量化目标检测模型:CornerNet-Ghost。采用特征提取网络Hourglass-Ghost作为骨干网络,对待测物体的左上和右下角点进行检测,并搭配级联角点池化优化提取的角点位置。实验结果表明,CornerNet-Ghost性能超过现有主流的轻量级角点检测网络CornerNet-Squeeze,且在检测计算时间远少于大型网络的条件下达到相近的准确性。
At present,there are many target detection tasks in industry,but the conventional large-scale target detection network can not run smoothly due to equipment reasons,and there is a large demand for lightweight target detection network.Aimed at this problem,a new lightweight target detection model,CornerNet-Ghost,is proposed.The feature extraction network Hourglass-Ghost was used as the backbone network to detect the upper left and lower right corners of the object to be measured.The cascaded corner pooling was used to optimize the extracted corner position.The experimental results show that the performance of CornerNe-Ghost is better than that of CornerNet-Squeeze,which is the mainstream lightweight corner detection network,and achieves similar accuracy when the detection and calculation time is far less than that of large network.
作者
张莲
余松林
Zhang Lian;Yu Songlin(School of Electrical and Electronic Engineering,Chongqing University of Technology,Chongqing 400054,China)
出处
《计算机应用与软件》
北大核心
2023年第9期236-241,共6页
Computer Applications and Software
基金
重庆市教委项目(KJQN201801142)。
关键词
轻量级网络
幽灵
反残差结构
中继监督
级联角点池化
Lightweight network
Ghost
Inverted residual
Intermediate supervision
Cascade corner pooling
