Research On Key Technologies Of Scene Matching Areas Selection Of Cruise Missile

The scene matching areas selection is a key technology of scene matching aided navigation.The research on this technology has important practical significance to speed up the missile test and appraisal process and save the test expenditure.In this paper,the scene matching aided navigation were taken as the research background.The features of scene matching aided navigation were combined with the characteristics of the multi-sensor scene matching.The matching features of scene matching area were built and the technology of scene matching based on machine learning and the selection of scene matching area based on multi feature fusion were studied in this dissertation.Experiments shows that the two methods both can be applied to automatically select the scene matching area,which can meet the navigation requirements of scene matching.In this dissertation,matching probability and matching accuracy,the two basic suitability evaluation indexes for mathing suitability,were firstly difined and discussed.In most of the cases,there are no corresponding image pairs applied for scene matching area selecting.The calculation method of self matching probability of single image was studied in detail.The quality decline method of single image and the calculation method of self matching probability were defined as the basis for the follow-up study of sence matching areas selection algorithm.In matching accuracy analysis,the definition of matching accuracy based on error mean and the definition of matching accuracy based on error variance were discussed.The two definitions both can effectively reflect the matching accuracy obtained by multiple sets of tests.Next,combining the engineering application,two algorithms of sence matching areas selection were studied.The one is the evaluation method of matching suitability analysis based on hierarchical screening was studied.Firstly,the number of edge points of each image area is counted and the image areas of which the number of edge points is below the threshold will be eliminated,so it can greatly reduce the number of candidate areas and save the time of calculation.Secondly,the LSSVRM model is used to predict the matching probability of the candidate areas selected in the first step,most of the areas with lower prediction matching probability can be eliminated in this step.As the input of the LSSVRM model,the matching suitability characteristics of image areas can directly affect the training effect of the model.Two matching suitability characteristics presented in this paper and ten matching suitability characteristics presented in the past research articles were used to built area feature description vector.The vector has 12 dimensions.By analyzing the correlation of the feature index,the number of dimensions of the vector is reduced from 12 dimensions to 9 dimensions.The 9 dimensions vector is used as the model input.The model test shows that the LSSVRM model can be used to predict the area’s matching probability,so as to realize the rapid screening of the scene area.Finally,the self matching probability of each scene area selected in the second step,which has a higher accuracy but needs a large amount of calculation,is calculated.The final matching area selection results is gotten by sorting the value of the self matching probability.The other one is the scene matching area selection method based on multi feature fusion.Firstly,the image feature insex set which can reflect the matching area selection criteria is built.Then the two fork tree structure model is used to fusion feature index and the best comprehensive feature are obtained by genetic algorithm.Finally,the comprehensive feature is used to achieve matching area selection.This method can effectively ensure the richness,significance,stability and uniqueness of the selected areas.In the last of this paper,based on the research of the two scene matching area selection methods and according to the scene matching area selection process,the scene matching area selection tests of the same scene with different matching parameters and different scenes with the same matching parameters were carried out.The tests show that both the two scene matching areas selection methods can bu used to select the scene matching areas which can meet the requirements of the scene matching aided navigation.And the results has a high level of consistency with the manual selection results.