Weather Image Recognition Considering Light Condition Via SENet for Intelligent Traffic System
Download as PDF
DOI: 10.23977/CNCI2020070
Author(s)
Yizhou Ding, Rui Ren and Chengxi Luo
Corresponding Author
Yizhou Ding
ABSTRACT
The comprehensive automatic classification of weather category and environmental light condition has important applications in the field of intelligent Traffic. This paper proposes a weather image classification model based on SENet, which can identify 5 types of weather: sunny, cloudy, rain, fog, and snow, and can evaluate the environmental brightness (Bright, Dark) of each type of weather above. This paper uses a weather image dataset consisting of 8890 weather images, of which 7900 are the training set and 990 are the validation set. A variety of neural networks (AlexNet, GoogLeNet, and BPNN) were used to train and compare the datasets. After comparison, it was found that the accuracy of SENet was as high as 98.48% when the epoch was 200. The accuracy of extreme weather recognition is higher than the other three networks, which is more suitable for extreme weather recognition.
KEYWORDS
Extreme weather recognition; light condition evaluation; SENet; classification; intelligent traffic system introduction