Scene image classification method based on Alex-Net model
Jing Sun, Xibiao Cai, Fuming Sun, Jianguo Zhang · 2016
Deep convolutional neural network (DCNN) is a powerful method of learning image features with more discriminative and has been studied deeply and applied widely in the field of computer vision and pattern recognition. In order to further explore the superior performance of DCNN and improve the accuracy of the scene image classification, this paper presents a novel algorithm of scene classification, which fully learning the deep characteristics of the images based on the classical Alex-Net model and support vector machine. In the first place, we use the Alex-Net model learning scene image features and extract the last layer with 4096 neurons of the Alex-Net model as the image features in this method; Then, we use the Lib-SVM training model for scene image classification and compare with classification method based on the regression model; Finally, we carried out the experiments on two common datasets in this paper. The experimental results have shown that DCNN can extract the image features effectively. Meanwhile, the trained scene model also has stronger generalization performance and achieves the state-of-the-art classification accuracy.