Detection and recognition of military vehicle from a given image or a video frame with the help of unmanned aircraft system is the major issue which we are concerned about. Vehicle identification and classification from a resource constraint device embedded on an aerial vehicle integrated with an intelligent object detection algorithm, is a big support for defence agency. The vehicle can be controlled both manually and autonomously. However, there is no military objects/vehicles dataset openly available with different varieties of military classes. Hence, we propose our dataset having 6772 images with classes namely, Military Trucks, Military Tanks, Military Aircrafts, Military Helicopters, Civilian Car and Civilian aircraft. Quantize SSD Mobilenet v2 and Tiny Yolo v3 deep learning models are trained on our dataset and compared its performance over resources constraint edge devices. The observations and results from the research show that Tiny Yolo v3 performs well over the other model and is highly efficient and can even run with edge based devices due to it’s light weight. There is a detailed generalised mathematical calculation provided which calculates the number of flight paths and the total number of frames required to cover a given area for surveillance using available hardware specification. This work will be suitable for classifying the military and civilian vehicle in the real time scenario using edge device over UAVs.

在无人机系统的帮助下，从给定的图像或视频帧中检测和识别军用车辆是我们关注的主要问题。嵌入在飞行器上的资源约束装置与智能目标检测算法相结合，对车辆进行识别和分类，是国防机构的一大支持。车辆可以手动和自动控制。但是，没有公开可用的不同军事类别的军事物体/车辆数据集。因此，我们提出我们的数据集有 6772 张图像，包括军用卡车、军用坦克、军用飞机、军用直升机、民用汽车和民用飞机。Quantize SSD Mobilenet v2 和 Tiny Yolo v3 深度学习模型在我们的数据集上进行了训练，并比较了其相对于资源受限边缘设备的性能。研究的观察和结果表明，Tiny Yolo v3 的性能优于其他模型，并且效率很高，由于重量轻，甚至可以与基于边缘的设备一起运行。提供了详细的广义数学计算，该计算使用可用的硬件规范来计算覆盖给定监视区域所需的飞行路径数和帧总数。这项工作将适用于在无人机上使用边缘设备在实时场景中对军用和民用车辆进行分类。研究的观察和结果表明，Tiny Yolo v3 的性能优于其他模型，并且效率很高，由于重量轻，甚至可以与基于边缘的设备一起运行。提供了详细的广义数学计算，该计算使用可用的硬件规范来计算覆盖给定监视区域所需的飞行路径数和帧总数。这项工作将适用于在无人机上使用边缘设备在实时场景中对军用和民用车辆进行分类。研究的观察和结果表明，Tiny Yolo v3 的性能优于其他模型，并且效率很高，由于重量轻，甚至可以与基于边缘的设备一起运行。提供了详细的广义数学计算，该计算使用可用的硬件规范来计算覆盖给定监视区域所需的飞行路径数和帧总数。这项工作将适用于在无人机上使用边缘设备在实时场景中对军用和民用车辆进行分类。提供了详细的广义数学计算，该计算使用可用的硬件规范来计算覆盖给定监视区域所需的飞行路径数和帧总数。这项工作将适用于在无人机上使用边缘设备在实时场景中对军用和民用车辆进行分类。提供了详细的广义数学计算，该计算使用可用的硬件规范来计算覆盖给定监视区域所需的飞行路径数和帧总数。这项工作将适用于在无人机上使用边缘设备在实时场景中对军用和民用车辆进行分类。