Study On The Detection Of The Small And Dim Target In Magnesium Melt Based On Image Processing Technology

This paper was depended on the National Natural Science Foundation of China “The automatic detection method and experimental study of hydrogen content in magnesium melt based on real-time image processing”(51374007). Detection of small and dim targets under complex background has been one of the hot points of image processing field, and magnesium melt images are very complex and the first bubble belongs to scope of small and dim targets. The exact recognition is the key link of detection and calculating of hydrogen content, meanwhile, it is also a significant procedure to ensure high-quality magnesium alloy castings.Based on the analysis of characteristics of magnesium melt images, the basic theories of wavelet transform and image fusion were researched in the aspect of image fusion of wavelet transform, then a detection method was proposed, and the results of the test showed that the method could successfully recognize the target. A detection method of the first bubble based on support vector machine was proposed aiming at the drawbacks of BP neural net, finally, it can detect the bubble combining with improved OSTU. The test demonstrated that the proposed method had better performance in the respects of SNR,gain of SNR, and real-time.Different factors of image complexity were adopted to respectively analyze complexity of global background and neighborhood background corresponding to bubbles of different places precipitating, aiming at researching on influence of random precipitation of the first bubble on image detection methods, finally, the universal applicability of image detection methods, which was about precipitation of the first bubble, was explored. Based on comparison between the first bubble and scum, characteristic difference of the first bubble and scum was analyzed. Then, a simple method of real-time identification was established, and the results showed that the method could preliminarily recognize detection result by using existing data to experiment.The two proposed image detection methods aiming at the detection of the first bubble can successfully detect the target. What's more, some important achievements in the aspects of random precipitation of the first bubble and real-time distinguish. In a word,these are very significant to further explore detection methods of hydrogen content in magnesium melt.