For hyperspectral remote sensing image (HSI) classification, the learning process of deep neural networks has been progressively advanced in depth, but the fine features are often largely lost or even disappear in the process of depth transfer. With the increase in feature aggregation and connectivity, the complexity of the network and the training parameters increases greatly, requiring more training time. This paper proposed a multi-scale dense network (MSDN) for HSI classification that made full use of different scale information in the network structure and combined scale information throughout the network. It implemented feature extraction of HSIs in two dimensions, including the features at fine and coarse levels. In the horizontal direction, it considered the deep extraction of HSI features, and the 3-D dense connection structure was used for aggregating features at different levels. In the vertical direction, scale information was considered, and three-scale feature maps at low, middle, and high levels were generated based on the first layer of the network. The MSDN used stride convolution for downsampling and combined feature information at different scale levels. The MSDN extracted features along the diagonal line. The network implemented the reconstruction of deep feature extraction and multi-scale fusion for HSI classification. The MSDN model performed well on representative HSI datasets, namely, the Indian Pines, Pavia University, Salinas, Botswana, and Kennedy Space Center datasets. It improved the training speed and accuracy for HSI classification and especially improved the convergence speed, which effectively saved computing resources and had high stability.

This study introduces the attention mechanism in hyperspectral remote sensing image (HSI) classification which can strengthen the information provided by important features, and weaken the non-essential information. We introduced the Squeeze-and-Excitation (SE) block embedded in three-dimensional densely connected convolutional network (3D-DenseNet) to form 3D-SE-DenseNet for HSI classifications. This model can learn a powerful network with low training costs and fast convergence speed, and avoids overfitting on small sample datasets. Two different 3D-SE-DenseNet models of 3D-SE-DenseNet and 3D-SE-DenseNet-BC were set up. Results from experiments show that the 3D-SE-DenseNet performs well on the Indian Pines, Pavia University, Botswana, and Kennedy Space Centre datasets.

Deep learning models have shown excellent performance in the hyperspectral remote sensing image (HSI) classification. In particular, convolutional neural networks (CNNs) have received widespread attention because of their powerful feature-extraction ability. Recently, a capsule network (CapsNet) was introduced to boost the performance of CNNs, marking a remarkable progress in the field of HSI classification. In this article, we propose a novel deep convolutional capsule neural network (DC-CapsNet) based on spectral-spatial features to improve the performance of CapsNet in the HSI classification while significantly reducing the computation cost of the model. Specifically, a convolutional capsule layer based on the extension of dynamic routing using 3-D convolution is used to reduce the number of parameters and enhance the robustness of the learned spectral-spatial features. Furthermore, a lighter and stronger decoder network composed of deconvolutional layers as a better regularization term and capable of acquiring more spatial relationships is used to further improve the HSI classification accuracy with low computation cost. In this study, we tested the performance of the proposed model on four widely used HSI datasets: the Kennedy Space Center, Indian Pines, Pavia University, and Salinas datasets. We found that the DC-CapsNet achieved high classification accuracy with limited training samples and effectively reduced the computation cost.

Hyperspectral remote sensing images (HSIs) are rich in spectral-spatial information. The deep learning models can help to automatically extract and discover these rich information from HSIs for classifying HSIs. However, the sampling of the models and the design of the hyperparameters depend on the number of samples and the size of each sample's input space. In the case of limited samples, the description dimension of features is also limited and overfitting to other remote sensing image datasets is evident. This study proposes a novel deep feature aggregation network (DFAN) for HSI classification based on a 3D-convolutional neural network (3D-CNN) from the perspective of feature aggregation patterns. By introducing the residual learning and dense connectivity strategies, we established a deep feature residual network (DFRN) and a deep feature dense network (DFDN) to exploit the low-, middle- and high-level features in HSIs. For the Indian Pines and Kennedy Space Center datasets, the DFRN model was determined to be more accurate. On the Pavia University dataset, both the DFDN and DFRN have basically the same accuracy, but the DFDN has faster convergence speed and more stable performance on the validation set than the DFRN. Therefore, when faced with different HSI data, the corresponding aggregation method can be chosen more flexibly according to the requirements on number of training samples and the convergence speed. This is beneficial in HSI classification.

In this study, we introduce a non-local block of the attention mechanism into capsule neural network (CapsNet) to form a non-local capsule network (NLCapsNet) for hyperspectral remote sensing image (HSI) classification. The presented NLCapsNet uses global information from input images and has a powerful representation of the capacity and spatial relationships among HSI features. It can effectively isolate invalid information and consolidate valid information, in addition to learning more representative features and capturing the long-distance dependencies of HSIs with only a few layers. An additional convolutional layer is embedded before the capsule layers to capture high-level features and speed up the routing procedure. The proposed method can effectively enhance the classification accuracy with a rapid convergence speed and avoid overfitting when the number of training samples is limited. The NLCapsNet performs well on the classification of the Kennedy Space Center, Pavia University and Salinas datasets.

The influence of tree distribution and slope on canopy gap fraction (GF) and bidirectional reflectance factor (BRF) is shown here to be non-negligible. Trees are often assumed to be randomly distributed in natural forests due to random distribution of natural resources, but this assumption is not valid for forest plantations. A geometric optical model for forest plantations (GOFP) is a geometric optical model for forest plantations on horizontal surfaces based on the theory of exclusion distance among crowns. Sloping terrains change the exclusion distance among crowns, and inevitably affect the canopy GF and BRF. In this article, GOFP with a hypergeometric model (distances among trees are considered) on horizontal surfaces is modified as GOFP-T to simulate BRF for forest plantations on sloping terrains under two scenarios: (horizontal distances among crowns remain unchanged with slope) and (sloping distances among crowns remain unchanged with slope). Two three-dimensional (3-D) radiative transfer models (DART and LESS) and field measurements are used to evaluate and validate GOFP-T simulations. The results show that 1) the canopy GF, four component area ratios, and canopy BRF simulated by GOFP-T show high consistency with results from the two 3-D model: root-mean-square errors in GF, sunlit foliage, and sunlit ground are less than 0.02, 0.06, and 0.03, respectively; 2) forest coverage and canopy reflectance in GOFP-T are compared well with point cloud results from an airboard LiDAR system and Landsat 8 OLI surface reflectance products, respectively, indicating that GOFP-T has ability in simulating canopy reflectance for forest plantations on sloping terrains. GOFP-T with the hypergeometric model in this article is the first simple model to simulate canopy GF and BRF for forest plantations on sloping terrains.