The WGΓ model has been validated as an effective model for the characteristic of polarimetric synthetic aperture radar (PolSAR) data statistics. However, due to the complexity of natural scene and the influence of coherent wave, the WGΓ model still needs to be improved to fully consider the polarimetric information. Then, we propose the WGΓ mixture model (WGΓMM) for PolSAR data to maintain the correlations among statistics in PolSAR data. To further consider the spatial-contextual information in PolSAR image classification, we propose a novel mixture model, named mixture WGΓ-Markov random field (MWGΓMRF) model, by introducing the MRF to improve the WGΓMM model for classification. In each law of the MWGΓ-MRF model, the interaction term based on the edge penalty function is constructed by the edge-based multilevel-logistic model, while the likelihood term being constructed by the WGΓ model, so that each law of the MWGΓ-MRF model can achieve an energy function and has its contribution to the inference of attributive class. Then, the mixture energy function of the MWGΓ-MRF model has the fusion of the weighted component, given the energy functions of every law. The mixture coefficient and the corresponding mean covariance matrix of the MWGΓ-MRF model are estimated by the expectation-maximization algorithm, while the parameters of the WGΓ model being estimated by the method of matrix log-cumulants. Experiments on simulated data and real PolSAR images demonstrate the effectiveness of the MWGΓ-MRF model and illustrate that it can provide strong noise immunity, get smoother homogeneous areas, and obtain more accurate edge locations.

In this article, we propose a hybrid conditional random field model based on complex-valued 3-D convolutional neural network (C3NN), named as HCRF-C3NN, for polarimetric synthetic aperture radar (PolSAR) image classification. HCRF-C3NN combines the modeling power of random fields with the representation-learning ability of deep learning model for PolSAR image. By C3NN, HCRF-C3NN exploits deep amplitude and phase information of PolSAR data and, thus, measures the class probabilities in the framework of random fields. Additionally, HCRF-C3NN captures the spatial label interactions effectively in PolSAR image classification by the following work: 1) Based on the C3NN-based class probabilities, the relative entropy of class distributions is derived to describe the local structure and, thus, regulates the pairwise label interactions, which enhances the accuracy of edge location in the classification; 2) a product-of-expert potential is introduced to enforce label consistency, and, thus, HCRF-C3NN is more robust against speckle of PolSAR image and achieves smoother classification within the consistent class region. At last, to capture PolSAR image information in a more complete manner, the deep features and PolSAR scattering statistics are integrated into HCRF-C3NN based on Bayesian fusion. In this way, HCRF-C3NN effectively explores spatial correlation, scattering statistics, and deep features of PolSAR data in the framework of random fields. The experimental results demonstrate the superiority of HCRF-C3NN over the recent deep learning models for PolSAR image classification.

To effectively fuse the high-dimensional features, in this letter, we propose the composite kernel and hybrid discriminative random field model, abbreviated as CK-hybrid discriminative random field (HDRF), for polarimetric synthetic aperture radar (PolSAR) image classification. In the CK-HDRF model, given high-dimensional features with different characteristics, the unary potential is constructed by relating multiple kernel k-means (MKKM) clustering to the traditional HDRF model. In this way, the high-dimensional decomposition and texture features can be well fused, thus making their deserved contributions to the inference of the attributive class and further increasing the discrimination capacity of CK-HDRF. The pairwise potential is constructed by the generalized Ising model with an additional edge penalty function, and thus, it can well capture the underlying spatial relationship and maintain the edge locations in classification. Moreover, the statistics of PolSAR data are modeled by the Wishart-generalized gamma (WGΓ ) distribution. Experiments on real PolSAR images demonstrate the effectiveness of CK-HDRF in classification.

Fusing features from different feature descriptors or different convolutional layers can improve the of understanding of scene and enhance the classification accuracy. In this paper, we propose a hierarchical deep texture feature fusion network, abbreviated as HDTFF-Net, aiming to improve the classification accuracy of high-resolution remote sensing scene classification (RSSC). The proposed HDTFF-Net can effectively combine the shallow texture information from manual features and the deep texture information by convolutional neural networks (CNNs). First, for deeply excavating the multi-scale and multi-directional shallow texture features in images, an improved Wavelet feature extraction module and a Gabor feature extraction module are designed by fully fusing the structural features into the backbone neural network. Then, to make the output texture features more discriminative and interpretative, we incorporate the above texture feature extraction modules into traditional convolutional neural networks, and design two improved deep networks, namely Wave-CNN and Gabor-CNN. Finally, according to the Dempster- Shafer (D-S) evidence theory, the designed Wave-CNN and Gabor-CNN are fused with the traditional CNN (Tra-CNN) by a decision-level fusion strategy, which can effectively capture the deep texture features by different feature descriptors and improve the classification performance. Experiments on high-resolution remote sensing images demonstrate the effectiveness of the proposed HDTFF-Net, and verify that it can greatly improve the classification performance.

The remote sensing scene classification (RSSC) based on convolutional neural networks (CNNs) are generally limited by the complex background interference and the difficulty of identifying key targets in image. Thus, this letter proposes a semantic-aware fusion network for RSSC, abbreviated as SAF-Net, to better construct discriminative features and effectively fuse features for classification. The proposed SAF-Net, which employs the ResNet50 pretrained on the ImageNet dataset as the backbone network, mainly contains the semantic-aware module and the multilayer feature fusion module (MFFM). The semantic-aware module utilizes a spatial enhanced module (SEM) and a covariance channel attention module (CCAM) to accurately capture the discriminative semantic features. It can precisely identify and extract the essential semantic elements in image, such as distinct object types and their spatial distributions. Then, the MFFM uses the features learned by the semantic-aware module to guide other layers for effective feature fusion through a self-attention mechanism. It can not only enriches the feature representation of SAF-Net but also ensure the effective fusion of the semantic information. Extensive comparisons and ablation experiments on remote sensing datasets demonstrate the effectiveness of the proposed SAF-Net, and verify that it can greatly improve the classification performance.