适用于电表终端故障识别训练的图像训练方法  

Image training Method Suitable for Fault Recognition of Smart Meter Terminals

作　　者：丁超 张秋雁 王蓝苓 欧家祥 王铎润 DING Chao;ZHANG Qiuyan;WANG Lanling;OU Jiaxiang;WANG Duorun(Electric Power Research Institute of Guizhou Power Grid Co.,Ltd.,Guiyang 562400,China;Guiyang Xiuwen Power Supply Bureau of Guizhou Power Grid Co.,Ltd.,Guiyang 550200,China)

机构地区：[1]贵州电网有限责任公司电力科学研究院,贵州贵阳562400 [2]贵州电网有限责任公司贵阳修文供电局,贵州贵阳550200

出　　处：《电工技术》2020年第12期56-59,共4页Electric Engineering

摘　　要：针对当前深度学习算法在电表故障识别训练领域中存在的不足,文章提出了一种改进的电表故障识别训练方法。对传统算法的识别训练过程进行了优化,重点关注缺陷样本的训练和图像采集质量的优化,并采用透视变换等技术手段丰富样本库、调整改善样本采集质量,提升了数据训练结果的可靠性。实例验证表明该方法具有良好适应性与高识别训练可靠度。In view of the shortcomings of the current deep learning algorithm in the field of intelligent meter fault recognition,an improved intelligent meter fault recognition method was proposed.The recognition process of the traditional algorithm was optimized,focusing on the training of the defect samples and the optimization of the image acquisition quality.The techniques such as perspective transformation was used to enrich the sample library.The sample collection quality,and greatly improve the reliability of the data training results were adjusted and improved.The example verification shows the method has good adaptability and high recognition reliability.

关 键 词：电表终端 深度学习 样本不平衡 故障识别训练 图像训练 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]