基于深度学习的变压器图像识别系统

针对变压器型号多、图像复杂,以及传统基于机器学习的人工设计特征的方法不能对大规模变压器图像准确分类等问题提出了基于深度学习的变压器图像识别系统直接对原始图像进行"端对端"的学习。为实现变压器图像的准确分类,提出了改进VGG-16卷积神经网络的变压器图像识别模型。在VGG-16模型的基础上,重新构建了全连接层,针对原有的SoftMax分类器,采用3标签的SoftMax分类器进行替换,以实现网络结构优化,并通过迁移学习共享V GG-16模型卷积层和降采样层的权值参数。通过构建变压器图像的训练集和测试集对改进模型进行了训练,并进行性能测试。结果表明,与深度神经网络、卷积神经网络模型相比,改进VGG-16模型具有更好的效果,识别误差达到了9.17%,并实现了对3种变压器的准确区分。

Transformer Image Recognition System Based on Deep Learning

Aiming at the traditional methods based on machine learning artificial design features which can not accurately classify large-scale transformer images because transformers have many types and their images are complex,a transformer image recognition system is proposed based on deep learning,which directly performs "end-to-end" learning on the original image.In order to achieve accurate classification of transformer images,a transformer image recognition model based on improved VGG-16 convolutional neural network is proposed.Based on the VGG-16 convolutional neural network model,the fully connected layer is reconstructed.The original SoftMax classifier is replaced with the 3-label classifier to optimize the network structure.The weighting parameters of the shared VGG-16 model convolutional layer and down-sampling layer are learned by transfer.By building a transformer image training set and test set,the improved model training and the performance of the improved method are tested.The experimental results show that compared with the deep neural network and the convolutional neural network model,the improved VGG-16 has better effect,achieving a recognition error of 9.17%,and achieving accurate differentiation of the three transformers.