Deep Convolution Neural Network Based Research on Recognition of Mine Vehicle Head and Tail

The mine environmental monitoring system captures the photos of the head and the tail of the vehicle, and sometimes the system can not accurately distinct whether it is the head or the tail of the vehicle. When there are two trucks in the view of the surveillance camera, the captured image contains the head of one truck and the tail of another truck. What needs to be recognized is the head license plate number or the tail license plate number. However, because the system cannot distinguish the head and tail of the truck, it will cause more false alarms. In order to solve this problem, this paper proposes an end-to-end feature extraction and recognition model based on deep convolution neural network (Deep CNN). The Deep CNN model contains five stage CNN layer and each layer contains different kernel size to extract the features. The data set is provided by Huaibei Siyuan Technology Co., Ltd., which includes normal capture, escape and false alarm images of the trucks. The final prediction rate is 85% on the testing set, which occupied twenty percent of the whole image set. The prediction rate of our model has been higher than the prediction rate base on right-out-left-in principle, which is used in the mine environmental monitoring system. Finally, our model will be applied in the mine environmental monitoring system.