基于改进Faster R-CNN的变压器巡检图像数据挖掘研究

Research on Image Data Mining of Transformer Inspection Based on Improved FAST R-CNN

随着我国电网智能化、信息化的建设与发展,电网中的电力设备通过长期的运维、检修和试验,积累了大量的各种形式的电力数据。其中,相比于主要以数值形式存储的结构化数据而言,电力设备巡检图像等非结构化数据,具有更广泛的应用场景和更高的价值密度,但由于不能被计算机直接识别和处理,其挖掘过程也存在更多的难点。本文提出基于改进Faster R-CNN模型的电力设备图像目标检测方法。以主变压器的巡检图像为例,考虑了主变压器各个部件的尺寸差异较大以及部件位置之间存在关联性的特点,对Faster R-CNN模型的结构进行了改进,有效提高了主变压器多部件类别和位置识别的准确率,为识别不同部件的缺陷和故障现象奠定了基础。

With the construction and development of China's power grid intelligence and informatization,the power equipment in the power grid has accumulated a large number of various forms of power data through long-term operation and maintenance,maintenance and test.Compared with the structured data mainly stored in numerical form,unstructured data such as inspection images of power equipment have more extensive application scenarios and higher value density.However,because it can not be directly recognized and processed by computer,the mining process also has more diff iculties.In this paper,a target detection method based on improved Faster R-CNN model for power equipment image is proposed.Taking the inspection image of main transformer as an example,considering the large size difference of each component of main transformer and the correlation between component positions,the structure of Faster R-CNN model is improved,which effectively improves the accuracy of multi component classif ication and location identif ication of main transformer,and lays a foundation for identif ying defects and fault phenomena of different components.