关键词: 模式识别, 超声波钢轨探伤, 支持向量机, DSP实时信号处理

Abstract:

This paper describes a more efficient real time SVM(support vector machine)-based ultrasonic rail defect detection and classification system. Feature extraction is achieved based on the attribute of ultrasonic rail defect and then SVM classification prediction algorithm and statistical processing are used to realize classification and calculating the size of the rail defect. This machine learning algorithm is tested in DSP and the type, grade and location of the defects are displayed in real-time.

Key words: pattern recognition, ultrasonic rail flaw detection, support vector machine(SVM), DSP real-time signal processing