In recent years, convolution neural networks (CNNs) have been widely used in the field of remote sensing scene image classification. However, CNN models with good classification performance tend to have high complexity, and CNN models with low complexity are difficult to obtain high classification accuracy. These models hardly achieve a good trade-off between classification accuracy and model complexity. To solve this problem, we made the following three improvements and proposed a lightweight modular network model. First, we proposed a lightweight self-compensated convolution (SCC). Although traditional convolution can effectively extract features from the input feature map, when there are a large number of filters (such as 512 or 1024 common filters), this process takes a long time. To speed up the network without increasing the computational load, we proposed a self-compensated convolution. The core idea of this convolution is to perform traditional convolution by reducing the number of filters, and then compensate the convoluted channels by input features. It incorporates shallow features into the deep and complex features, which helps to improve the speed and classification accuracy of the model. In addition, we proposed a self-compensating bottleneck module (SCBM) based on the self-compensating convolution. The wider channel shortcut in this module facilitates more shallow information to be transferred to the deeper layer and improves the feature extraction ability of the model. Finally, we used the proposed self-compensation bottleneck module to construct a lightweight and modular self-compensation convolution neural network (SCCNN) for remote sensing scene image classification. The network is built by reusing bottleneck modules with the same structure. A lot of experiments were carried out on six open and challenging remote sensing image scene datasets. The experimental results show that the classification performance of the proposed method is superior to some of the state-of-the-art classification methods with less parameters.

For remote sensing scene image classification, many convolution neural networks improve the classification accuracy at the cost of the time and space complexity of the models. This leads to a slow running speed for the model and cannot realize a trade-off between the model accuracy and the model running speed. As the network deepens, it is difficult to extract the key features with a sample double branched structure, and it also leads to the loss of shallow features, which is unfavorable to the classification of remote sensing scene images. To solve this problem, we propose a dual branch multi-level feature dense fusion-based lightweight convolutional neural network (BMDF-LCNN). The network structure can fully extract the information of the current layer through 3 × 3 depthwise separable convolution and 1 × 1 standard convolution, identity branches, and fuse with the features extracted from the previous layer 1 × 1 standard convolution, thus avoiding the loss of shallow information due to network deepening. In addition, we propose a downsampling structure that is more suitable for extracting the shallow features of the network by using the pooled branch to downsample and the convolution branch to compensate for the pooled features. Experiments were carried out on four open and challenging remote sensing image scene data sets. The experimental results show that the proposed method has higher classification accuracy and lower model complexity than some state-of-the-art classification methods and realizes the trade-off between model accuracy and model running speed.

With the development of computer vision, attention mechanisms have been widely studied. Although the introduction of an attention module into a network model can help to improve classification performance on remote sensing scene images, the direct introduction of an attention module can increase the number of model parameters and amount of calculation, resulting in slower model operations. To solve this problem, we carried out the following work. First, a channel attention module and spatial attention module were constructed. The input features were enhanced through channel attention and spatial attention separately, and the features recalibrated by the attention modules were fused to obtain the features with hybrid attention. Then, to reduce the increase in parameters caused by the attention module, a group-wise hybrid attention module was constructed. The group-wise hybrid attention module divided the input features into four groups along the channel dimension, then used the hybrid attention mechanism to enhance the features in the channel and spatial dimensions for each group, then fused the features of the four groups along the channel dimension. Through the use of the group-wise hybrid attention module, the number of parameters and computational burden of the network were greatly reduced, and the running time of the network was shortened. Finally, a lightweight convolutional neural network was constructed based on the group-wise hybrid attention (LCNN-GWHA) for remote sensing scene image classification. Experiments on four open and challenging remote sensing scene datasets demonstrated that the proposed method has great advantages, in terms of classification accuracy, even with a very low number of parameters.

Convolutional neural networks are widely used in the field of hyperspectral image classification. After continuous exploration and research in recent years, convolutional neural networks have achieved good classification performance in the field of hyperspectral image classification. However, we have to face two main challenges that restrict the improvement of hyperspectral classification accuracy, namely, the high dimension of hyperspectral images and the small number of training samples. In order to solve these problems, in this paper, a new hyperspectral classification method is proposed. First, a three-dimensional octave convolution (3D-OCONV) is proposed. Subsequently, a dense connection structure of three-dimensional asymmetric convolution (DC-TAC) is designed. In the spectral branch, the spectral features are extracted through a combination of the 3D-OCONV and spectral attention modules, followed by the DC-TAC. In the spatial branch, a three-dimensional, multiscale spatial attention module (3D-MSSAM) is presented. The spatial information is fully extracted using the 3D-OCONV, 3D-MSSAM, and DC-TAC. Finally, the spectral and spatial information extracted from the two branches is fully fused with an interactive information fusion module. Compared to some state-of-the-art classification methods, the proposed method shows superior classification performance with a small number of training samples on four public datasets.

In recent years, convolutional neural networks (CNNs) have been widely used for hyperspectral image classification, which show good performance. Compared with using sufficient training samples for classification, the classification accuracy of hyperspectral images is easily affected by a small number of samples. Moreover, although CNNs can effectively classify hyperspectral images, due to the rich spatial and spectral information of hyperspectral images, the efficiency of feature extraction still needs to be further improved. In order to solve these problems, a spatial–spectral attention fusion network using four branch multiscale block (FBMB) to extract spectral features and 3D-Softpool to extract spatial features is proposed. The network consists of three main parts. These three parts are connected in turn to fully extract the features of hyperspectral images. In the first part, four different branches are used to fully extract spectral features. The convolution kernel size of each branch is different. Spectral attention block is adopted behind each branch. In the second part, the spectral features are reused through dense connection blocks, and then the spectral attention module is utilized to refine the extracted spectral features. In the third part, it mainly extracts spatial features. The DenseNet module and spatial attention block jointly extract spatial features. The spatial features are fused with the previously extracted spectral features. Experiments are carried out on four commonly used hyperspectral data sets. The experimental results show that the proposed method has better classification performance than some existing classification methods when using a small number of training samples.