磁粉检测图像智能辨识方法研究及实现-模式识别与智能系统专业论文.docx

西南科技大学硕士研究生学位论文第 西南科技大学硕士研究生学位论文 第 PAGE IV 页 西南科技大学硕士研究生学位论文第 西南科技大学硕士研究生学位论文 第 III 页 Abstract With the development of modern industry, the traditional industry is now upgrading towards intelligent automation. As an important method of nondestructive testing (NDT), magnetic particle testing becomes an urgent need to improve the automation. To meet the requirements of magnetic particle testing image intelligent recognition method, this paper carried out the theoretical and practical studies closely. It mainly includes the following aspects: First, combining with the work process of the semi automated fluorescent magnetic particle testing machine, we briefly analyze the anti-interference techniques of the image acquisition system . Second, through the analysis of the characteristics of magnetic particle testing images, we designed an algorithm to preprocess and segment image based on mapping relationship from the light wavelength to RGB. Then we analyze and extract crack image features combining the magnetic powder detecting process, including area characteristics, circumference characteristics, length characteristics, length-width ratio characteristics, characteristics of the circular degree, and degree of rectangular characteristics. Furthermore, in the classification stage, we analyze and study the classification algorithm, and propose a classification and mapping approach combining with support vector machine (SVM) classification of posterior probability, and achieved intelligent automatic classification of the crack magnetic particle detection images. Finally, experimental analysis has proved the effectiveness and practicality of the proposed approach.. Regarding to the two important indicators in automatic image recognition of magnetic particle testing, the missing rate and false alarm rate are greatly improved. This paper realizes the automatic classification of magnetic particle crack image based on semi automatic magnetic par