基于深度学习模型小型化技术的输电线路智能巡检研究与应用

摘要:

为了解决目前输电线路智能巡检图像采集真空期长以及效率低的问题,提出了基于深度学习模型小型化技术的电力输电线路智能巡检方案。通过飞桨的模型压缩库Paddle Slim,将可以进行输电线路异常识别的模型小型化;再将小型化模型部署在端侧推理引擎Paddle Lite驱动的无源无线的移动设备中,实现了在低算力、低功耗情况下的输电线路智能巡检。方案成果在某电网公司中应用实践,验证了方案的实用性。

Abstract:

A intelligent inspection method for transmission lines based on deep learning model miniaturization is proposed to solve the long vacuum period and low efficiency of image acquisition in the current intelligent inspection of transmission lines.Using the Paddle Slim,the model miniaturization tools from Paddle Paddle deep learning framework can compress the model of identification abnormal situation around transmission lines. Then the compressed model is deployed in wireless mobile devices driven by the edge-computing inference engine Paddle Lite. The proposed solution is applied in a power grid utility to verify its practicability.