Improved image classification algorithm based on convolutional neural network

Abstract

At present, convolutional neural networks are widely used in image classification, but the training effect of the network model with a relatively shallow number of layers is not good, and the model with a deeper network is prone to overfitting problems at the end. This article uses the cat and dog data set, and the selection is relatively mature The VGG16 model was improved. Add a dropout layer and a feature extraction layer to it, and perform L2 regularization on the loss function at the end to deepen the model depth and improve the fit of the entire model. The experimental results show that the improved model can greatly improve the detection accuracy.