摘 要 :目的 细粒度军事目标数据集是实现现代战争目标自动分类的重要支撑数据之一。当前缺乏高质量的细粒度军事目标遥感图像数据集，制约了军事目标自动精准检测的研究。本文收集并标注了一个新的军用车辆细粒度检测的遥感图像数据集，并基于此设计了一种基于YOLOv5s的改进模型来提高军用车辆目标检测性能。方法 该数据集来源于谷歌地球数据，收集了亚洲、北美洲和欧洲范围内40多个军事场景下的3000张遥感图像，包含国外多个国家和地区的军用车辆目标。经高质量人工水平边界框标注，最终形成包含5个类别共计32626个实例的军用车辆细粒度检测遥感图像数据集。针对遥感图像中军用车辆识别难题，本文提出的基准模型考虑了遥感军用车辆目标较小、形状和外观较为模糊以及类间相似性、类内差异性大的特点，基于YOLOv5s设计了基于目标尺寸的跨尺度检测头和上下文聚合模块，提升细粒度军用车辆目标的检测性能。结果 提出的基准模型在军用车辆细粒度检测遥感图像数据集上的实验表明，对比经典的目标检测模型，新基准模型在平均精度均值（mean average precision，mAP）指标上提高了1.1%。结论 本文构建的军用车辆细粒度检测遥感图像数据集为军事目标自动分类算法的研究提供了参考与支持，有助于更为全面地研究遥感图像中军用车辆目标的特性。

Abstract: Objective Informational warfare has put forward higher requirements for military reconnaissance, and military target identification, as one of the main tasks of military reconnaissance, needs to be able to deal with fine-grained military targets and provide personnel with more detailed target information. Optical remote sensing image datasets play a crucial role in remote sensing target detection tasks, which can provide valuable standard remote sensing data for model training, and also provide objective and uniform benchmarks for comparison of different networks and algorithms. The current lack of high-quality fine-grained military target remote sensing image datasets constrains the research on automatic and accurate detection of military targets. Informational warfare has put forward higher requirements for military reconnaissance, and military target identification, as one of the main tasks of military reconnaissance, needs to be able to deal with fine-grained military targets and provide personnel with more detailed target information. Optical remote sensing image datasets play a crucial role in remote sensing target detection tasks, which can provide valuable standard remote sensing data for model training, and also provide objective and uniform benchmarks for comparison of different networks and algorithms. As a special remote sensing target, military vehicles also have characteristics such as environmental camouflage, shape and structure changes, and movement "color shadows", which make the detection task more challenging. Figure 1 shows the challenges posed by the fine-grained target characteristics of military vehicles in optical remote sensing images, which can be categorized into five types according to the source of target characteristics. 1) Target characterization as affected by satellite remote sensing imaging systems. 2) Characterization of the vehicle target itself. 3) Military Vehicle Target Characterization. 4) Characteristics affected by the combination .5) Properties of fine-grained classification. Translated with www.DeepL.com/Translator (free version)The current lack of high-quality fine-grained military target remote sensing image datasets constrains the research on automatic and accurate detection of military targets. In order to promote the development of deep learning-based research on fine-grained accurate detection of military vehicles in high-resolution remote sensing images, this paper constructs a new high-resolution optical remote sensing image dataset MVRSD (Military Vehicle Remote Sensing Dataset), and based on it, we design an improved model based on YOLOv5s to improve the target detection performance of military vehicles. Method The dataset is derived from Google Earth data, collected 3,000 remotely sensed images from more than 40 military scenarios within Asia, North America and Europe, and 32,626 military vehicle targets were acquired from the remotely sensed images. The spatial resolution of the image is 0.3m and the size is 640×640. The dataset consists of remotely sensed images and corresponding labeled files, and the targets were manually selected and classified by experts through the interpretation of high-resolution optical images. The granular categories in the dataset are divided into five categories based on vehicle size and military function: Small Military Vehicles (SMV), Large Military Vehicles (LMV), Armored Fighting Vehicles (AFV), Military Construction Vehicles (MCV), and Civilian Vehicles (CV). Military Construction Vehicles (MCV), Civilian Vehicles (CV). The geographic environments of the samples include cities, plains, mountains and deserts. Aiming at the difficulty of recognizing military vehicles in remote sensing images, the benchmark model proposed in this paper takes into account the characteristics of remote sensing military vehicles with smaller targets, fuzzy shapes and appearances, as well as interclass similarity and intraclass variability. The number of instances of each category in the dataset and the number of instances in each image depend on their actual distribution in the remote sensing scene, which can reflect the realism and challenge of the dataset. A cross-scale target size-based detection head and context aggregation module are designed based on YOLOv5s to improve the detection performance of fine-grained military vehicle targets. Result In this paper, the characteristics of military vehicle targets in remote sensing images and the challenges facing the fine-grained detection of military vehicles are analyzed, and for the problem that the YOLOv5 algorithm does not have high detection accuracy for small targets and is prone to omission and misdetection, an improved model based on YOLOv5s is proposed as a baseline model, and a cross-scale detector head is selected based on the dimensions of the targets in the dataset in order to efficiently detect the targets at different scales, and at the same time the attention mechanism module is inserted in front of the detector head in order to inhibit the interference of complex backgrounds on the target. Based on this dataset, five target detection models were applied for experiments. Experiments of the proposed benchmark model on a remote sensing image dataset for fine-grained detection of military vehicles show that the new benchmark model improves the mean average precision (MAP) metric by 1.1% compared to the classical target detection model. The experimental results show that the deep learning model based on the dataset can achieve good results in fine-grained accurate detection of military vehicles. Conclusion The MVRSD dataset can support researchers to analyze the features of remote sensing images of military vehicles from different countries, and provide training and test data for deep learning methods, which will provide dataset support for the study of fine-grained detection of military vehicles. It is shown that the proposed benchmark method can effectively improve the target detection accuracy of remotely sensed military vehicles. The MVRSD dataset is available at https://github.com/baidongls/MVRSD.