Multi-modal fusion traffic information classification algorithm based on distributed optical fiber sensing

The realization of intelligent transportation is inseparable from the perception and processing of traffic information. In the field of intelligent transportation, video surveillance, infrared, magnetic induction sensors, other methods have been applied, to a certain extent to meet the needs. However, the above method has some limitations, such as complex installation, high cost, blind spot detection, weather influence. Therefore, this paper proposes a traffic information classification algorithm based on distributed optical fiber sensing, which uses the distributed perception ability of optical fiber to obtain traffic vibration data, combined with video data, and realizes high-precision classification of vibration signals through neural network. This method uses multi-modal fusion technology to extract key features of fiber distributed sensing data and video data respectively through ResNet and video understanding network, and then extracts multi-scale features corresponding to the two modes from different levels of the feature extraction network. Finally, multi-modal cross-attention feature enhancement fusion module is used to achieve multi-modal feature fusion. The information complementation between different modes is realized effectively, and the classification of traffic information is completed. In this study, we conducted tests on the traffic data set we collected, and the framework showed excellent performance in various types of identification, which can provide a reference for the construction of intelligent transportation.