ERF-YOLO: A YOLO algorithm compatible with fewer parameters and higher accuracy
作者：
Highlights：
• Research on the effective receptive field and analyze the factors that increase the effective receptive field.
• Based on the YOLO v4 backbone network, optimize the activation function to increase the effective receptive field.
• Use Conv_dw instead of an ordinary convolutional network to reduce the number of parameters.
• Increase the number of short-circuiting and stacking, improve features and strengthen network performance.
• Optimize the regression loss function of the anchor frame.
摘要
•Research on the effective receptive field and analyze the factors that increase the effective receptive field.•Based on the YOLO v4 backbone network, optimize the activation function to increase the effective receptive field.•Use Conv_dw instead of an ordinary convolutional network to reduce the number of parameters.•Increase the number of short-circuiting and stacking, improve features and strengthen network performance.•Optimize the regression loss function of the anchor frame.
论文关键词：The effective receptive field,The activation function,The backbone network,Concat,The anchor box loss function
论文评审过程：Received 6 September 2021, Revised 28 September 2021, Accepted 30 September 2021, Available online 20 October 2021, Version of Record 5 November 2021.
论文官网地址：https://doi.org/10.1016/j.imavis.2021.104317
