Method for classifying grid equipment based on convolution neural network

Abstract

The present invention discloses a method for classifying grid equipment based on the convolution neural network. The method comprises that Step 1, a convolution neural network model is trained according to the existing training set and test set of the grid equipment image, and the input layer preprocesses the input image data, so as to increase the amount of data, wherein the number of convolution layers is not greater than N, and N+1 is the number of convolution layers of the common convolution neural network; and Step 2, the trained convolution neural network model is used to carry out classification on grid equipment images that need to be classified. According to the method disclosed by the present invention, the data enhancement technology is used to preprocess the input image data to increase the amount of data, so that the problems that the network is over-fitted and the precision is decreased due to that the amount of data is insufficient are solved; and considering that the amount of the trained data is relatively small, the number of the convolution layers and the number of convolution kernels are reduced while the size of the convolution kernel is increased and the size of the feature map extracted by each layer of the convolution layers is reduced, so that the number of features extracted by the convolution layers is reduced, the effect of preventing the over-fitted phenomenon is reached, and accuracy is improved.

Description

technical field [0001] The invention relates to a method for classifying power grid equipment based on a convolutional neural network, which belongs to the field of neural networks. Background technique [0002] Grid equipment identification has very important applications in the fields of grid equipment classification, status monitoring and abnormal warning, and is a technology with high practical value. [0003] In recent years, there have been many breakthroughs in image recognition methods based on deep convolutional neural networks. However, due to the limitation of the amount of image data and the limitation of CPU computing power, the accuracy of the neural network has been difficult to break through, and the training efficiency is very low. With the implementation of data augmentation techniques and computing using GPUs, it becomes possible to achieve accurate classification of images using deep convolutional networks based on less data. [0004] At present, the ma...

AI Technical Summary

Problems solved by technology

[0008] The disadvantage is that for less training data, too many convolutional layers and convolution kernels in Alexnet are very prone to data overfitting, making the trained network very inaccurate and impractical for the classification results of test data.

Images