Research On Multi-sensor Exhaust Images Registration Based On Mutual Information

In order to solve the problem of difficult monitoring and management of vehicle and ship exhaust pollution at present,a remote sensing on-line monitoring method of exhaust is put forward by wuhan Institute of Physics and Mathematics.Because multi-sensor is used to collect the exhaust images in the system,the registration becomes a key link.The quality of registration results is directly related to the accuracy of exhaust monitoring results,so a reasonable and effective method is needed to solve the problem of exhaust image registration in the on-line monitoring system.At present,there are many researches on registration direction,and this technique is also widely used in medicine,remote sensing and computer vision.However,the research on the registration of multi-sensor exhaust images in the system is very rare.The traditional method can not satisfy the needs of the on-line exhaust detection system,so it is of great significance to do deep research on the registration of multi-sensor exhaust images.In this paper,the existing registration algorithms were studied and improved based on the characteristics of multi-sensor exhaust images,and an algorithm suitable for the registration of multi-sensor exhaust images was obtained.Then the algorithm was implemented to meet the engineering requirement.The details of the study were as follows:(1)In this paper traditional mutual information and SIFT operator which represent two different ways of thinking,were used to carry out preliminary registration experiments on multi-sensor exhaust images,and a conclusion was drawn that the traditional mutual information method was more suitable for multi-sensor exhaust images registration.But the experimental results also showed that the registration accuracy of traditional mutual information was not enough.(2)In order to improve the registration accuracy of exhaust image,the optimization search strategy of traditional mutual information method was deeply studied in this paper.Based on the existing optimization methods,a double-scale search genetic algorithm was proposed.Compared with the existing algorithms in the experiment,it showed that the double-scale search genetic algorithm had the advantage of high searching precision and this algorithm could overcome the local extremum.(3)Because the effective pixel ratio of exhaust image was small and there were many interference factors,the traditional mutual information used as similarity measure could not well describe the similarity of different exhaust images.The existing metrics were optimized and combined in this paper,and the multiplicative mixed mutual information and the additive mixed mutual information were proposed to describe the similarity between the exhaust images.The comparison experiment was completed with different similarity measures and the results showed that the additive mixed mutual information could describe the similarity between exhaust images under different circumstances more perfectly.(4)According to the need of engineering,the software module of image registration for multi-sensor exhaust images based on QT was designed and realized,and the corresponding functions were tested.Finally,a set of inversion experiments of sulfur dioxide concentration in exhaust images before and after registration were carried out to verify the effectiveness of the proposed exhaust image registration algorithm in this paper.