基于GoogLeNet Inception-V3模型的电力设备图像识别

Image Recognition of Electric Equipment Based on GoogLeNet Inception-V3 Model

随着电网智能运检的不断推进,电力设备状态监测产生了海量图像数据,然而目前尚无十分有效的方法对其进行智能分类识别。为解决机器学习对图像中的复杂特征提取困难和常规卷积神经网络学习能力不足带来的数据堆积和误判等问题,提出一种基于GoogLeNet Inception-V3模型的电力设备图像识别方法。首先介绍电力设备图像识别的模型结构及实现步骤,随后阐述该模型在特征提取中的高效性和准确性,最后通过实验证明该方法的优势。研究结果表明,提出方法对断路器、电流互感器、绝缘子、避雷器和电压互感器的平均识别准确率高达92.0%,比浅层CNN、k NN分类算法、VGG-16、GoogLeNet Inception-V1模型分别高32.5%、24.0%、6.5%和4.0%,具有较高的可行性和工程实用价值。

With the development of intelligent operation and inspection of power grid,massive image data has been generated by condition monitoring of electric equipment,but there is currently no very effective method for intelligent classification and recognition.In order to solve the problems of data accumulation and misjudgment caused by the difficulty of extracting complex features in images and the lack of learning ability of convolutional neural networks,an image recognition method for electric equipment based on GoogLeNet Inception-V3 model is presented.Firstly,introduce the model structure and implementation steps of electric equipment image recognition,then explain the efficiency and accuracy of the model in feature extraction,and finally prove the advantages of the method through experiments.The research results show that the proposed method has an average recognition accuracy of92% for circuit breakers,current transformers,voltage transformers,insulators and arresters,which is better than the shallow CNN,kNN,VGG-16,GoogLeNet Inception-V1 model is 32.5%,24.0%,6.5% and 4.0% respectively,it has the high feasibility and practical value in engineering.