关键词: 卷积神经网络（CNN）, Inception模块, Maxout网络, dropout操作, 遥感图像分类

Abstract: The sparse network structure for traditional Convolutional Neural Network (CNN) can not preserve the high efficiency of dense network-intensive computing and the empirical selection of the activation function in the experiment process. The results are inaccurate or increase the computational complexity In this paper, an improved convolution neural network method is proposed to classify remote sensing images.The Inception module and Maxout network (after dropout) combined with CNN, strengthen the ability to extract the characteristic of convolution operation, the network can fit any activation function at the same time to a certain extent to avoid overfitting effects. Experiment on the same US Land Use Classification Data Set (UCM-LandUse), The classification accuracy of the proposed method is more than 15% higher than that of the low-level feature classification method, which is about 10% higher than that of the visual dictionary based on the middle feature. It is about 4% higher than the traditional convolution neural network method. It can be seen that the proposed method has a high efficiency of remote sensing image classification.

Key words: CNN, Inception module, Maxout network, Dropout operation, Remote sensing image classification