An image classification method and device

Abstract

The invention discloses an image classification method and device. Based on a large image data set, the AlexNet model structure is trained; five trained convolutional layers are transferred to a small database to form a bottom layer feature extraction layer, which is combined with a two-layer convolutional layer. The residual network layer, multi-scale pooling layer, feature layer and softmax classifier are constructed to obtain the migration model structure; the image small data set is input into the migration model structure, and the batch gradient descent method is used to update the parameters to train the image classification hybrid model; according to The image classification hybrid model performs classification and obtains classification results. This application transfers the pre-trained convolutional layers on the large data set to the small data set, adds a multi-scale pooling layer, and connects the feature quantities output by the residual network layer and the multi-scale pooling layer into the classification The device increases the amount of features and alleviates the problem of overfitting; and the hybrid model trained based on convolutional neural network and transfer learning can effectively improve the accuracy of image classification.

Description

technical field [0001] The present invention relates to the field of transfer learning and deep learning, in particular to an image classification method and device. Background technique [0002] Convolutional Neural Networks (CNN) is an efficient recognition method. Generally, the basic structure of CNN includes two layers, one is the feature extraction layer, the input of each neuron is connected to the local receptive field of the previous layer, and the local features are extracted. Once the local feature is extracted, the positional relationship between it and other features is also determined; the second is the feature map layer, each calculation layer of the network is composed of multiple feature maps, each feature map is a plane, All neurons on the plane have equal weights. [0003] Transfer learning is the influence of one kind of learning on another kind of learning, which widely exists in the learning of knowledge, skills, attitudes and behavioral norms. Any k...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide an image classification method and device, the purpose of which is to solve the problem of image classification rate caused by over-fitting phenomenon and insufficient image feature extraction in the prior art when using small data sets to train image classification models

Images