军用光缆网是重要的**基础通信设施，传统的人工徒步巡检是查找光缆线路隐患的主要措施，但其耗时长,人力物力消耗大，易受敷设方式和地形环境变化影响。而采用无人机进行光缆线路巡检，时效性强,安全性高且经济性好，是未来的重点发展方向。由于工程车辆施工挖掘是造成光缆线路障碍的最主要原因，为此，文章提出将深度学习更快的基于区域的卷积神经网络(Faster R-CNN)目标检测方法应用到无人机航拍巡检图像的工程车辆检测中。基于航空影像中的车辆检测(VEDAI)公共数据集制作了工程车辆数据集，通过仿真训练和测试，实现了航拍图像中挖掘机和推土机等工程车辆的Faster R-CNN目标检测，检测平均精度(AP)值达0.659，优于传统的可变形组建模型(DPM)和方向梯度直方图+局部二值模式+支持向量机(HOG+LBP+SVM)等机器学习检测算法，研究结果可为军用光缆线路的无人机巡检应用研究提供一定的参考。

The military optical cable network is an important communication infrastructure for army. The traditional artificial walking patrol and inspection is the main method to find the hidden dangers of the optical cable lines. However, it takes a long time, and consumes a lot of manpower and material resources. It is also vulnerable to the changes in laying methods and terrain environment. However, the use of unmanned aerial vehicle for the patrol and inspection of optical cable lines, is the main development direction in the future due to the characteristics of fast, safe and economical. Because the factor of excavation in construction is the most important reason for the obstacle of the optical cable lines, it is proposed to apply the deep learning Faster Region-based Convolutional Neural Network (Faster R-CNN) target detection method to detect the image of engineering vehicle obtained by unmanned aerial vehicle. Based on the Vehicle Detection in Aerial Imagery (VEDAI) public data set, the engineering vehicle data set is made. Through simulation, the target detection of engineering vehicles such as excavators and bulldozers in aerial photography is achieved. The Average Precision(AP) value can reach 0.659. The detection results are better than traditional Deformable Part Model(DPM) and Histogram of Oriented Gradient + Local Binary Pattern + Support Vector Machine (HOG+LBP+SVM) machine learning detection methods. The study can provide a certain reference for the application of patrol and inspection for military optical cable lines by unmanned aerial vehicle.