Machine-vision detection for rail-steel’s surface flaws based on quantum neural network

Conventional detecting methods bring disadvantages of low-efficiency or high-fallout rate as for rail-steelpsilas surface flaws because of its non-planar and sophisticated contour. A journal machine vision approach was presented, in which imaging method and classifier algorithm are illustrated. Liner CCD is adapting to imaging for moving rail-steel. The classifier based on quantum neutral network (QNN) algorithm could deal with those similar and hardly differentiated ROI of flaws. It discussed feature vector parameters extracted from different spaces, moreover, QNNpsilas model, multi-level motivation functions based on Sigmoid function and training algorithm are expatiated in detail. An experimental device was developed and test results demonstrate the feasibility of the detection approach. It has proved the effectiveness and value of proposed method in automatic detection for rail-steelpsilas surface flaws.