基于深度卷积神经网络和多核学习的遥感图像分类方法

为解决传统遥感图像分类方法特征提取过程复杂、特征表现力不强等问题，该文提出一种基于深度卷积神经网络和多核学习的高分辨率遥感图像分类方法。首先基于深度卷积神经网络对遥感图像数据集进行训练，学习得到两个全连接层的输出将作为遥感图像的两种高层特征；然后采用多核学习理论训练适合这两种高层特征的核函数，并将它们映射到高维空间，实现两种高层特征在高维空间的自适应融合；最后在多核融合特征的基础上，设计一种基于多核学习-支持向量机的遥感图像分类器，对遥感图像进行精确分类。实验结果表明，与目前已有的基于深度学习的遥感图像分类方法相比，该算法在分类准确率、误分类率和Kappa系数等性能指标上均有所提升，在实验测试集上3个指标分别达到了96.43%, 3.57%和96.25%，取得了令人满意的结果。

Remote Sensing Image Classification Method Based on Deep Convolution Neural Network and Multi-kernel Learning

To solve the problems of complex feature extraction process and low characteristic expressiveness of traditional remote sensing image classification methods, a high resolution remote sensing image classification method based on deep convolution neural network and multi-kernel learning is proposed. Firstly, the deep convolution neural network is constructed to train the remote sensing image data set to learn the outputs of two fully connected layers, which are taken as two high-level features of remote sensing images. Then, the multi-kernel learning is used to train the kernel functions for these two high-level features, so that they can be mapped to the high dimensional space, where these two features are fused adaptively. Finally, with the combined features, a remote sensing image classifier based on Multi-Kernel Learning-Support Vector Machine (MKL-SVM) is designed for remote sensing image classification. Experimental results show that compared with the existing deep learning based remote sensing classification methods, the proposed algorithm achieves improved results in terms of classification accuracy, error, and Kappa coefficient. On the experimental test set, the above three indicators reach 96.43%, 3.57%, and 96.25% respectively, and satisfactory results are obtained.