Research On The Segmentation And Retrieval Of Rotary Kiln Flame Image

Recently, with the development of automatic rotary, computer hardware, software, and digital image processing technology, the computer has been replacing the people to watch the fire in the production process of rotary kiln. This would generate a mass of flame digital image data, and it was becoming a much more important issue that how to search and use these images effectively.The basic idea of the content-based image retrieval was used to query the various characteristics which described the images, and then returned the proximate results after calculating the similarity of the feature values. Now, the issue has being a research focus in image processing, which was widely applied in some fields, such as medical image data retrieval, face database, transport database and so on. In this paper a further analysis on the rotary kiln flame images was presented, and some experiments, which included image preprocessing, image segmenting and feature extraction, had been done. After this, the image database system was established, image retrieval and clustering analysis based on the system was introduced. The main contents and contributions are as follows:Firstly, the image processing technology of the rotary kiln control system was introduced. The research and discussion of key technology and application about clustering in the content-based image retrieval were presented, and the gray system theory was reviewed and discussed.Secondly, an improved image segmentation algorithm based on genetic algorithm and Otsu theory was proposed. In contrast with the algorithm of single common threshold, general genetic algorithm or Otsu theory, the result of the improved one showed that the split was more effective, and the algorithm could be used effectively in practice. After the flame image being partitioned, we extracted the characteristics of flame, materials and texture. On the basis of these solutions, the flame image database system was designed and implemented.Thirdly, the gray association analysis based on data mining, which could solute the problems of similarity measurement, was introduced into the content-based image retrieval (CBIR). By the novel CBIR method, the features of the flame image were analyzed, the similarities between the queried images and the images in database were calculated, which based on the weight factors that derived from the gray association rules, and as a consequence, several adjacent flame images were retrieved. Finally, the image retrieval system was designed, and the simulation platform was introduced briefly, which provided a simulation environment for image retrieval algorithm.