基于自适应patch的Transformer检测小金具缺陷

Adaptive Patch-base Transformer for Detecting Small Metal Defects

输电线路中连接各类电力设备的小金具容易受到外部或人为因素影响而脱落、安装不到位,给电网安全稳定的运行带来了极大的安全隐患,因此及时准确定位和识别出线路中的缺陷极其重要。然而,这类器件在高空中尺度极小,分布太广,给缺陷检测带来了极大的挑战。将NLP(自然语言处理)中Transformer架构引入电力图像缺陷检测,并在输入上做了自适应分割改进,设计了一种新的Transformer,命名为自适应patch的Transformer(APT)。还建立了一个多场景多目标的小金具缺陷数据集,为检测输电线路小金具缺陷提供了基础。

Due to the influence of external or human factors,the equipment connected to various power equipment in the transmission line is easy to fall off and not be installed in place,which brings great security risks to the safe and stable operation of the power grid.Therefore,it is extremely important to locate and identify the defects in the line timely and accurately.However,these equipments are extremely small and widely distributed at high altitude,which brings great challenges to defect detection.In this paper,Transformer architecture in NLP(natural language processing)is introduced into power image defect detection and adaptive segmentation is improved on the input.A new Transformer named adaptive patch Transformer(APT)is designed.This paper also builds a multi-scene and multi-object data set of small hardware defects to provide base for detecting small hardware defects in transmission lines.