基于多任务学习的变电站设备腐蚀检测

Corrosion Detection of Substation Equipment Based on Multi-task Learning

摘要:

变电站设备的工作状态直接影响到电力系统的稳定性、可靠性、安全性及抵抗事故的能力。而在众多变电站设备故障原因中，金属材料腐蚀是十分常见的一种。于是提出了一种基于深度学习的变电站腐蚀检测方法，可以直接通过图像检测变电站设备的腐蚀位置以及腐蚀程度，有效完成变电站设备的巡视和实时监控工作，及时排除腐蚀风险，从而保障变电站稳定运行。为了验证所提出方法的有效性，采用经过裁剪、旋转、镜像扩充后的变电站设备腐蚀数据集对网络进行训练。最终结果表明，所提出的网络能够准确地对变电站设备进行腐蚀区域检测以及腐蚀程度检测。

Abstract:

The operational status of substation equipment directly influences the stability, reliability, safety and ability against accidents of power grid. Among numerous causes of substation equipment faults, corrosion of metal materials is a common issue. A method for substation corrosion detection based on deep learning is proposed, which can directly detect the corrosion location and corrosion degree in substation equipment through image detection, which can effectively accomplish the inspection and real-time monitoring of substation equipment, promptly eliminate potential hazards and ensure the stable operation of substation. In order to validate the effectiveness of the proposed method, the corrosion data set of substation equipment, which has been augmented through cropping, rotation and mirroring, is used to train the network. The results show that the proposed network can accurately detect the corrosion areas and corrosion degree of substation equipment.