利用生成对抗网络构建配网绝缘子缺陷库的研究

Study on Construction of Distribution Network Insulator Defect Database by Using Generation Countermeasures Network

阐述配网绝缘子缺陷图像的特征、图像标注、数据训练,卷积神经网络对配网绝缘子设备的深度学习识别,提出利用生成对抗网络进行训练缺陷、生成模型,完成数据集合创建,标注流程创建,训练方法实施到最终缺陷库生成,最终建立关键指标、视觉模型、人工介入分析的验证方法。结果表明,生成对抗网络有助于配电网设备的缺陷库构建。

This paper describes the characteristics, image annotation, data training of the distribution network insulator defect image, and the deep learning and recognition of the distribution network insulator equipment by the convolution neural network. It proposes to use the generation confrontation network to train the defect, generate the model, complete the creation of data sets, create the labeling process, implement the training method to the final defect database generation, and finally establish the verification method of key indicators, visual model, and artificial intervention analysis. The results show that the generation of confrontation network is helpful to the construction of defect database of distribution network equipment.