Vehicle logo detection based on deep convolutional networks. (March 2021)

• Record Type: Journal Article
• Authors: Zhang, Junxing; Yang, Shuo; Bo, Chunjuan; Zhang, Zhiyuan
• Is Part Of: Computers & electrical engineering. Volume 90(2021)
• Publication Date: 2021-03
• Subjects: Vehicle Logo Detection -- Lightweight Networks -- Separable Convolution -- Deep Learning; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2021.107004
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlight: A exclusive Convolutional Neural Networks structure for the task of vehicle logo detection. A multi-stage training policy for small-scale object detection. A lightweight network structure with separable convolution for the real-time detection of vehicle logo. Achieving the task of vehicle logo detection in embedded devices. Abstract: For intelligent transportation systems, vehicle logo is an important part in the research of vehicle information recognition. Deep convolutional neural networks (CNNs) have been more reasonable and stronger than artificial selection in feature extraction and expression for targets. In our previous work, we found that the network of feature extraction and the training policy of detection have great effects on the accuracy of vehicle logo. For improving small-scale object detecting precision, we change the training policy. According to our previous work, we propose a lightweight network structure with separable convolution to improve the real-time character for vehicle logo while implement the method in embedded devices. The experiment proves that our model can effectively improve the detection accuracy of vehicle logo. Our training policy is valid method for small-scale objects. The lightweight network can solve the equilibrium problem of detecting precision and speed. Graphical abstract: Image, graphical abstract