未爆子弹药图像数据集构建方法及其关键技术研究

Research on the Construction Method and Key Technologies of Unexploded Submunition Image Dataset

随着计算机技术和机器视觉技术的迅速发展与应用,探索基于“人工智能+”模型的未爆子弹药搜寻技术受到了广泛关注。但是,由于未爆子弹药具有一定的危险性和受军事应用的特殊性影响,数据集构建是目前亟待解决的瓶颈问题。本文由此出发,分别论述了真实实物图像数据集和利用实物图片进行三维重建数据集的构建方法及流程,重点分析了两种数据集构建过程中的相关关键技术及其优缺点,并给出了一种利用多目相机采集目标图像和地理坐标信息,然后利用深度学习算法进行目标特征提取、生成三维点云和融合三维图像。试验结果表明,采用该方法构建的三维数据集可以有效解决未爆子弹药现有数据集数据量不足的问题,最后展望了数据集构建方法的未来发展方向。

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 extensive 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 construction process of the two data sets.A multi-camera is used to collect the target image and geographic coordinate information,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 unexploded submunitions.Finally,the future development direction of the data set construction method is prospected.