Automatic Fluorescent Magnetic Particle Testing Target Recognition In Image Processing Technology

Considering the limited recognition ability and high false alarm rate (FAR) in automated fluorescent magnetic particle inspection, this paper deals with the high recognition rate classification of crack by image processing.In this paper, Support Vector Machine (SVM) based on Statistical Learning Machine (STL) is investigated, especially the C-SVM which introduced a penalty factor C and is used as a discriminant model to detect the flaw in railway wheel surface. The main characteristics are described from four aspects:geometric, gray, gradient and texture features. Using the Gaussian kernel function, C-SVM search the parameters C, g based on cross validation (CV) to identify the crack. Sample selection and simplification methods are presented in the basis of skeleton of class blocks obtained by morphological grads and clustering algorithm automatically and quickly. This paper also offers a new method about rapid parameter optimization which introduces the GRAM matrix similarity method to get the best C, g in C-SVM discriminant model. Euclidean distance, vectorial angle cosine and correlation coefficient are combined as the new similarity measure and PSO numerical optimization algorithm is used to search the C-SVM parameters.The simulation results have proven that C-SVM can get a high rate of recognition low false alarm rate (LFAR) and good generalization performance in fluorescent magnetic particle inspection, and can get the sample easily and efficiently. The algorithm has a faster convergence and operation than CV.