WCSE

Algorithm Research on Vehicle Type Classification Based on Effi-YOLOX

Abstract— In order to improve the feature extraction and recognition capability of the model for vehicle images in different environments, this paper proposed a model named Effi-YOLOX to processing the vehicle type classification task. In this paper, the migration parameters based on Imagenet are used to initialize the EfficientNetv2 model, and the backbone of EfficientNetv2 is used as the backbone of Effi-YOLOX. Furthermore, MHSA(Multi-head Self-Attention) is used to replace SE-attention so that the model can extract more comprehensive features. On this basis, PANet+ is used to replace the PANet of the original YOLOX model; The Coordinate Attention(CoordAttenton) is introduced to make the model focus more on important features. The experimental results show that the Effi-YOLOX is superior to the existing classical models in the vehicle type classification task.

Index Terms— Vehicle Type Classification, Convolutional Neural Network, Transformer, Attention Mechanism

You Zhou
School of Automation, Xi'an University of Post and Telecommunications
Leilei Ma
School of Automation, Xi'an University of Post and Telecommunications

[Download]

Cite: You Zhou, Leilei Ma, " Algorithm Research on Vehicle Type Classification Based on Effi-YOLOX, " WCSE 2022 Spring Event: 2022 9th International Conference on Industrial Engineering and Applications, pp. 1471-1476, Sanya, China, April 15-18, 2022.