The adhesion state between the wheel/rail is applied as an input training weak classifier instead of the visual rail state in the thesis. After combination of the weak classifiers which have been trained, BP neural network strong classifier based on Adaboost algorithm is obtained. In the meantime, a strong predictor is established by the method of a strong classifier establishment. Based on Matlab simulation platform, BP Adaboost algorithm is employed to carry out the simulation experiment of orbital state recognition in the thesis. The strong predictor is applied to predict the adhesion state in the simulation experiment. With the predictive value being the input, the dynamic recognition of the surface state is finished by the application of strong classifier. Meanwhile, the traditional BP neural network algorithm is employed to identify the orbital state, and the comparison of both recognition results is made. The simulation results show that the BP Adaboost algorithm can reduce the amount of calculation, the amount of training as well as the time of training when compared to the traditional BP neural network algorithm, and it can also solve the compensating problems in the track state recognition.
[1]
M. Zhang,et al.
Real-time traffic sign detection and recognition for intelligent vehicle
,
2014,
2014 IEEE International Conference on Mechatronics and Automation.
[2]
Zainul Abdin Jaffery,et al.
Maximally Stable Extremal Region Marking-Based Railway Track Surface Defect Sensing
,
2016,
IEEE Sensors Journal.
[3]
T Ohyama.
SOME BASIC STUDIES ON THE INFLUENCE OF SURFACE CONTAMINATION ON ADHESION FORCE BETWEEN WHEEL AND RAIL AT HIGHER SPEEDS
,
1989
.
[4]
Jie Cao,et al.
Real-Time Target Detection and Recognition with Deep Convolutional Networks for Intelligent Visual Surveillance
,
2016,
2016 IEEE/ACM 9th International Conference on Utility and Cloud Computing (UCC).