Discriminative feature learning is the key to remote sensing scene classification. Previous research has found that most of the existing convolutional neural networks (CNN) focus on the global semantic features and ignore shallower features (low-level and middle-level features). This study proposes a novel Lie Group deep learning model for remote sensing scene classification to solve the above-mentioned challenges. Firstly, we extract shallower and higher-level features from images based on Lie Group machine learning (LGML) and deep learning to improve the feature representation ability of the model. In addition, a parallel dilated convolution, a kernel decomposition, and a Lie Group kernel function are adopted to reduce the model’s parameters to prevent model degradation and over-fitting caused by the deepening of the model. Then, the spatial attention mechanism can enhance local semantic features and suppress irrelevant feature information. Finally, feature-level fusion is adopted to reduce redundant features and improve computational performance, and cross-entropy loss function based on label smoothing is used to improve the classification accuracy of the model. Comparative experiments on three public and challenging large-scale remote-sensing datasets show that our model improves the discriminative ability of features and achieves competitive accuracy against other state-of-the-art methods.

The key problem of remote sensing image classification is to understand the semantic content of the images effectively. Up to now, most methods are to improve the accuracy of the final classification through the convolutional neural networks model. However, the structure of the whole model is complex and contains a large number of parameters. To address this problem, in this study, we proposed a lightweight method of the intrinsic mean within Lie groups, which can achieve high accuracy. We apply the computational advantage of Lie group manifold space and design the kernel function that is suitable for both Lie group sample and vector sample. Experiments are performed on two publicly available and challenging data sets [aerial image dataset (AID) and Northwestern Polytechnical University-REmote Sensing Image Scene Classification, contains 45 scene classes (NWPU-RESISC45)], and the results show that our method is more accurate than the state-of-the-art methods.

Remote sensing scene classification is used to label specific semantic categories for images. The current methods have achieved competitive performances, but they are only for Euclidean space samples. As a result, their representations are not robust for non-Euclidean space samples, which affects the classification accuracy. In this letter, we introduce the Lie group manifold into the traditional feature representation method and propose a novel intrinsic mean representation method within the Lie group. At the same time, the kernel function based on the sample of the Lie group is designed to further improve the robustness and accuracy of classification. In addition, our method achieves satisfactory performance on two public and challenging remote sensing data sets of UC Merced and NWPU-RESISC45.