Vehicle classification using a real-time convolutional structure based on DWT pooling layer and SE blocks

作者：

Highlights：

• Introducing a new lightweight DNN architecture for vehicle classification.

• Use of DWT as a new pooling method.

• Combining classic Conv layers with Squeeze & Excitation block for better performance.

• Real Time operation of the proposed network (42 ms on CPU and 0.276 ms on GPU).

• Better performance than AlexNet, InceptionV3, ResNet-50 and VGG19 on two datasets.

摘要

•Introducing a new lightweight DNN architecture for vehicle classification.•Use of DWT as a new pooling method.•Combining classic Conv layers with Squeeze & Excitation block for better performance.•Real Time operation of the proposed network (42 ms on CPU and 0.276 ms on GPU).•Better performance than AlexNet, InceptionV3, ResNet-50 and VGG19 on two datasets.

论文关键词：Vehicle classification,Convolutional,Deep learning,Haar wavelet,Squeeze and excitation block

论文评审过程：Received 22 June 2020, Revised 19 April 2021, Accepted 10 June 2021, Available online 15 June 2021, Version of Record 17 June 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.115420