Magneto-optical Image Enhancement By Pattern Recognition

With the ferromagnetic materials widely used in major equipment of industrial,national defense and others, timely and effective monitoring of its structural health has become an important issue. Due to fastness, visualization and high efficiency,magneto-optical imaging(MOI) technology is favored in the application of non-destructive testing to ferromagnetic materials. The principle of MOI is to convert the defect information that existed in the specimen into optical signal and then show it in an image. However, optical signal can easily be disturbed by instrumentation and detection environment, thus a large number of static interference characteristics are introduced into the final magneto-optical image, which limits the detection performance of MOI technology. In order to solve the problem, researchers have proposed some effective image enhancement algorithms, such as motion-based filtering, connected domain method, frame difference method, and so on. Especially motion-based filtering,has achieved very good results in suppressing static interference. However,motion-based filtering method cannot achieve a single point defect detection, because the method obtain data by scanning different areas of specimen. In order to improve the motion-based filtering method, this paper enhances the defect characteristics by acquiring the magneto-optical image data in single point and using the frame difference method. And because the frame difference method cannot achieve the desired effect,this paper researches the image processing method based on pattern recognition, which using feature space method to extract the defect characteristics.Firstly, the generate mechanism of the interference characteristics in the magneto-optical image is analyzed before suppressing these characteristics by image processing, while interference characteristics are divided into two categories: speckle interference and stripe-like interference. And then introduce the Brewster's law to optimize the magneto-optical imaging platform, so that it is optimal in hardware design to get the best raw data.We use low-frequency AC excitation to achieve collecting images in single point through obtaining background interference images when zero-crossing, and effective detection images when peak, and then use frame difference for processing. In this way,we can improve the motion-based filtering technology. Then slightly adjust analyzer to strengthen the background interference in images, and optimize the image processing by Image fusion. Finally, analysis the reason why the frame difference method cannot achieve desired result.Because of the rough result of the frame difference method used in the improved motion-based filtering method, the magneto-optical image enhancement algorithm based on pattern recognition is proposed to replace the frame difference method in the improved scheme. Firstly, a feature space that reflecting the interference feature in images is constructed by the principal component analysis method. And then optimize the space model in two aspects: the training data volume and the spatial dimension. The test image is then projected into the space, and the interference feature component of the test image is extracted after retrieved. Finally, enhancement of the defect feature is realized by making a difference between the test image and its interference component.Based on the image evaluation standard customized, the validity of the method is proved by experiment and comparing with the frame difference method.