Research On Fragment Image Detection And Matching Based On Artificial Intelligence

Secondary fragments are armor fragments and uncontrolled projectile fragments which fly forward at high speed behind the protective layer of the equipment after the ammunition strikes the armored target.The flying speed,size and dispersion of fragments are significant parameters for researching the combat power of warhead and evaluating the damage efficiency of target.High-speed photography is an important means to test flight parameters of fragments.Because of its advantages of accuracy,intuition and high reliability,it has been widely used in weapon system performance test.High-speed photographic image processing is the core of fragment group parameter testing,so it is of great essential to research the secondary fragment image processing method.It can provide necessary basis for ammunition design,sizing,inspection and power evaluation.The secondary fragment group sequence images of high-speed camera are taken as the research object.Through artificial intelligence method,the problem of sequence fragment group image detection,fragment feature extraction and fragment image matching are researched,and the evolution process of fragment group is formed.For fragment target detection,based on Faster R-CNN model,the detection results are obtained through three parts:Region Proposal Network,ROI pooling and fragment frame prediction.In the Region Proposal Network of Faster R-CNN,a model based on modular transformation is designed.The Non-Maximum Suppression algorithm of Region Proposal Network is improved,which makes the location of fragment objective candidate box more accurate.The detection ability of fragment images with partial overlap and occlusion is further improved.In fragment feature extraction,through the construction of Gaussian difference pyramid and the detection of spatial extremum,the main direction is searched after the feature points are screened.The scale and rotation invariant features of fragments with remarkable characterization ability are extracted.In fragment images multi-objective matching,based on KNN algorithm,KNN is improved by using multi-dimensional tree structure model.The matching results are obtained by constructing a multi-dimensional tree structure model,searching the nearest neighbor of the model and predicting the model.The improved algorithm reduces the dimension of feature matching and improves the accuracy of image matching.Through the research of detection,feature extraction and matching,the temporal and spatial distribution of the whole process evolution of secondary fragments is obtained.The velocity distribution of secondary fragments is calculated,and the distribution law of secondary fragments is preliminarily given.It is verified by experiments that multi-objective images of fragments in complex background.Compared with the Faster R-CNN algorithm,the detection accuracy of the improved algorithm has increased by 5.4%on average.The accuracy of fragment target matching algorithm increased by 14.4%on average.The given distribution law of secondary fragments has reference significance for the power evaluation of related ammunition.