With the rapid development and application of computer technology and machine vision technology,the exploration of unexploded submunition search technology based on"artificial intelligence+"model has received exten-sive attention.However,due to the danger of unexploded submunitions and the particularity of military applications,data set construction is a bottleneck problem that needs to be solved urgently.Based on this,the paper discusses the construction methods and processes of real physical image data sets and three-dimensional reconstruction data sets using physical images.It focuses on the analysis of the key technologies and their advantages and disadvantages in the con-struction process of the two data sets.A multi-camera is used to collect the target image and geographic coordinate in-formation,and then the deep learning algorithm is used to extract the target feature,generate the three-dimensional point cloud and fuse the three-dimensional image.The experimental results show that the three-dimensional data set constructed by this method can effectively solve the problem of insufficient data volume of the existing data set of unex-ploded submunitions.Finally,the future development direction of the data set construction method is prospected.
维普
万方数据