基于改进边缘算法的通信光缆设备智能检测技术研究

Research on intelligent detection technology of communication optical cable equipment based on improved edge algorithm

在无人机巡检电力通信光缆设备时将产生的视觉数据卸载到云端处理,通过边缘计算优化电力通信光缆设备的深度学习检测运算量。由ROSLink(robot operating system,ROS)从机器人操作系统提取电力通信光缆设备的数据,然后嵌入JSON进行序列化后将信息发送至云端,进而实现图像的云计算卸载。深度学习模块中DeepBrain云子系统启动云服务器的多个GPU,高速同步处理多架无人机的批量图像。实验表明,将视觉数据卸载至云端后电压降低率和功耗是机载GPU处理的50%,可延长无人机续航时间、缩短图像实时处理时间。

When the unmanned aerial vehicle inspects the power communication optical cable equipment,the gen⁃erated visual data is offloaded to the cloud for processing,and the deep learning detection calculation of the power communication optical cable equipment is optimized through edge computing.ROSLink(robot operating system,ROS)extracted the data of the Power communication optical cable equipment from the robot operating system and then embeded JSON for serialization and sent the information to the cloud,thereby achieving cloud computing offloading of the image.The DeepBrain cloud subsystem in the deep learning module activated multiple GPUs of the cloud server to process batch images of various drones simultaneously at high speed.Experiments showed that the voltage reduction rate and power consumption after offloading the visual data to the cloud were 50%of the onboard GPU processing,extending the UAV’s lifetime and shortening real-time image processing time.