Abstract
The technology of intelligent detection of military vehicle objects has already become the foundation for tasks related to intelligence and tracking of weapons and equipment, which is essential for situational awareness in modern intelligent warfare. Over the past two decades, numerous datasets of military vehicle images have been collected, primarily focusing on the classification of various categories of military vehicles. However, almost all these datasets are not publicly available, and the publicly available ones suffer from annotation quality issues. To address the lack of datasets and the quality of existing public datasets, we propose a specialized dataset based on our own collection of images, as well as publicly available ones. To develop methods for automatic detection of various categories of military vehicles and distinguishing them from civilian vehicles, we created a new dataset for military vehicle detection and classification (DMVDC). It consists of 5,899 images of military vehicles collected using three different methods: automated scraping, manual selection of image search results, and data augmentation techniques. To the best of our knowledge, our DMVDC [1] dataset is the only publicly available dataset of military vehicles that consider hidden and partially visible objects. This dataset will contribute to the development of computer vision models for detecting military vehicles.