摘要
基于卷积神经网络参数冗余较大的问题,提出一种基于模型压缩算法改进YOLOv4-tiny的车辆检测方法,以提高检测速度。首先,通过特征图矩阵的秩判断通道对网络模型的重要程度,对模型的通道剪枝,减少模型的参数数量;其次,对局部样本点进行采样提取,使用知识蒸馏算法,使模型的精度得到回升。实验结果表明,改进后的YOLOv4-tiny网络模型检测精度仅损失2.9%的情况下,模型参数减少了51.1%;部署在Jeston Nano设备上运行,每秒帧率FPS提升了93.7%。
To solve the problem that the parameters of the convolutional neural network model are redundant,a vehicle detection method based on model compression algorithm is proposed to improve YOLOv4-tiny to increase the detection speed.First,the importance of the channel to the model is judged by the rank of the feature map matrix,and the channel of the network model is pruned to reduce the number of parameters of the network model;Secondly,the local sample points are sampled and extracted,and the knowledge distillation algorithm is used to make the network model more efficient,accuracy is improved.The experimental results show that when the detection accuracy of the improved YOLOv4-tiny network model is only lost by 2.9%,the model parameters are reduced by 51.1%;When deployed on a Jeston Nano device,the FPS is increased by 93.7%.
作者
单铭琦
文峰
高文印
SHAN Mingqi;WEN Feng;GAO Wenyin(Shenyang Ligong University,Shenyang 110159,China)
出处
《沈阳理工大学学报》
CAS
2023年第2期36-42,共7页
Journal of Shenyang Ligong University
基金
辽宁省教育厅高等学校基本科研项目(面上青年人才项目)(LJKZ0267)。
关键词
卷积神经网络
模型压缩
通道剪枝
知识蒸馏
convolutional neural network
model compression
knowledge distillation
channel pruning
