Heterogeneous Network-Based Contrastive Learning Method for PolSAR Land Cover Classification

Polarimetric Synthetic Aperture Radar (PolSAR) image classification is an important research content in the field of remote sensing. However, the lack of effective labeled data and the serious class distribution shift problem between different datasets hinder the application of deep learning networks in PolSAR image classification. Therefore, this paper proposes a Bi-classifier adversarial network based on covariance difference matrix and Convolutional Block Attention Module (CBAM) to solve the cross-domain problem in PolSAR images. Firstly, since PolSAR image processing generally uses the covariance matrix or coherence matrix of a single target pixel to construct the data set, which cannot effectively use the local information of PolSAR data,Therefore, the PolSAR coherence difference matrix P is proposed as the description of the PolSAR image difference information. Secondly, due to the different importance of features between neighboring pixels, the CBAM mechanism is introduced in order to exploit the spatial information of PolSAR data. Finally, in order to solve the cross-domain problem of PolSAR images more effectively, a bi-classifier adversarial network based on covariance difference matrix and CBAM is proposed. In this paper, we conduct experiments on public datasets, using images of San FranciscoⅠ acquired in 2008 as the source domain, and images of San Francisco Ⅱ acquired in 1992 as the target domain data. The experimental results show that the proposed method can effectively solve the cross-domain problem of PolSAR images.

Polarimetric Synthetic Aperture Radar (PolSAR) images encompass valuable information that can facilitate extensive land cover interpretation and generate diverse output products. Extracting meaningful features from PolSAR data poses challenges distinct from those encountered in optical imagery. Deep Learning (DL) methods offer effective solutions for overcoming these challenges in PolSAR feature extraction. Convolutional Neural Networks (CNNs) play a crucial role in capturing PolSAR image characteristics by exploiting kernel capabilities to consider local information and the complex-valued nature of PolSAR data. In this study, a novel three-branch fusion of Complex-Valued CNN named (CV-ASDF2Net) is proposed for PolSAR image classification. To validate the performance of the proposed method, classification results are compared against multiple state-of-the-art approaches using the Airborne Synthetic Aperture Radar (AIRSAR) datasets of Flevoland, San Francisco, and ESAR Oberpfaffenhofen dataset. Moreover, quantitative and qualitative evaluation measures are conducted to assess the classification performance. The results indicate that the proposed approach achieves notable improvements in Overall Accuracy (OA), with enhancements of 1.30% and 0.80% for the AIRSAR datasets, and 0.50% for the ESAR dataset. However, the most remarkable performance of the CV-ASDF2Net model is observed with the Flevoland dataset; the model achieves an impressive OA of 96.01% with only a 1% sampling ratio. The source code is available at: https://github.com/mqalkhatib/CV-ASDF2Net

Classification of the polarimetric synthetic aperture radar (PolSAR) data is an important facet of synthetic aperture radar (SAR) image interpretation. In general, many polarimetric features can be extracted from PolSAR data. However, not all features are benefit for classification and too many polarimetric features result in the low classification efficiency. We introduce a method to deal with many polarimetric features and obtain both high classification accuracy and high classification efficiency. We first preprocess the PolSAR data to obtain objects, i.e., homogeneous regions. Next, we apply polarimetric feature processing to compute various polarimetric features of the PolSAR data. Based on the classification of the sample information, we then select the prominent polarimetric features using the data mining attributes of decision tree algorithm (DTA). Finally, we obtain classification results through the selected polarimetric features by DTA, the sample data and support vector machine (SVM). We verified the proposed method using the AirSAR L-Band PolSAR data. Our experiments indicate that the classification accuracy of the proposed method is equivalent to that of SVM and the computational efficiency is comparable to that of DTA. Thus, the proposed method integrates the advantages of both DTA and SVM.

ABSTRACTThis paper proposes a new algorithm, for polarimetric synthetic aperture radar (PolSAR) classification, based on a stacked auto-encoder and scattering energy. Previous approaches to PolSAR classification predominantly consider only the single pixel of distribution of the polarimetric data and scattering characteristics, and ignore other kinds of image features like the relationship of the local pixels. Besides, because of the complexities of PolSAR data, it is difficult to compute the derivatives that are needed for back-propagation in deep-learning classifiers. To overcome these difficulties, we propose a new approach that combines the scattering power and stacks sparse auto-encoder (Scattering SSAE) for PolSAR classification. Firstly, orientation compensation is used to compensate the polarization orientation angle, reducing the impact of polarimetric angle noise. Secondly, Freeman-Durden decomposition is adopted to extract three basic scattering powers: surface, double bounce and volume. Each PolSAR image pixel is transformed into these scattering powers, yielding a new kind of feature from the PolSAR data. Finally, using the three kinds of scattering power as inputs, we combine local spatial information using a patch-based approach, and use a deep learning architecture to achieve classification. We compare our method against several other state-of-the-art methods using ground-truthed test-data, and show that the Scattering SSAE method achieves higher accuracy than other methods on most categories.

Polarimetric synthetic aperture radar (PolSAR) image classification has achieved great progress, but there still exist some obstacles. On the one hand, a large amount of PolSAR data is captured. Nevertheless, most of them are not labeled with land cover categories, which cannot be fully utilized. On the other hand, annotating PolSAR images relies more on domain knowledge and manpower, which makes pixel-level annotation harder. To alleviate the above problems, by integrating contrastive learning and transformer, we propose a novel patch-level PolSAR image classification, i.e., two-staged contrastive learning and sub-patch attention based network (TCSPANet). Firstly, the two-staged contrastive learning based network (TCNet) is designed for learning the representation information of PolSAR images without supervision, and obtaining the discrimination and comparability for actual land covers. Then, resorting to transformer, we construct the sub-patch attention encoder (SPAE) for modelling the context within patch samples. For training the TCSPANet, two patch-level datasets are built up based on unsupervised and semi-supervised methods. When predicting, the classification algorithm, classifying or splitting, is put forward to realise non-overlapping and coarse-to-fine patch-level classification. The classification results of multi-PolSAR images with one trained model suggests that our proposed model is superior to the compared methods.

Polarimetric synthetic aperture radar (PolSAR) images classification is an important topic for PolSAR images understanding and interpretation. However, traditional pixel-based PolSAR image classification that takes image pixel as a processing unit cannot make full use of spatial information and, thus, may not obtain the satisfactory classification results. Hence, this letter proposed an attention-based multiscale sequential network for PolSAR images classification increasing the multiscale spatial information between pixels by way of spatial sequence. Specifically, the long short-term memory (LSTM) network is introduced to convert the time sequence into spatial sequence to extract the spatial features. Then, to obtain the more abundant spatial features and select more important spatial information, an attention-based multiscale spatial-enhanced LSTM (AMSE-LSTM) is proposed to enhance the relationship between pixel spatial information. Finally, a new mixed loss function is defined to improve the classification performance. Experimental results with two real PolSAR data show that compared with state-of-the-art methods, the proposed method can achieve a much better performance and overall classification accuracy.

This paper proposes a new polarimetric synthetic aperture radar (PolSAR) image classification method which uses optimal feature selection and convolutional neural network (CNN). Firstly, original image features are extracted by performing target decomposition on PolSAR data. Then, in order to reduce feature dimension and improve image classification accuracy, a group of optimal and high correlation property features are selected with the method of filtering and wrappering from the original one. Finally, the optimized CNN model is designed to implement PolSAR image classification. Experiments with measured data show that the algorithm can choose a more effective polarization feature set and achieve better classification results.

The phase difference, amplitude product, and amplitude ratio between two polarizations are important discriminators for terrain classification, which derives a significant statistical-distribution-based polarimetric synthetic aperture radar (PolSAR) image classification. Traditionally, statistical-distribution-based PolSAR image classification models pay attention to two aspects: searching for a suitable distribution to model certain PolSAR image and a satisfactory solution for the corresponding distribution model with samples in every terrain. Usually, the described distribution form is too complicated to build. Besides, inaccurate parameter estimation may lead to poor classification performance for PolSAR image. In order to refrain from this phenomenon, a variational thought is adopted for the statistical-distribution-based PolSAR classification method in this paper. First, a mixture Wishart model is built to model the PolSAR image to replace the complicated distribution for the PolSAR image. Second, a learning-based method is suggested instead of inaccurate point estimation of parameters to determine the distribution for every class in the mixture Wishart model. Finally, the proposed learning-based mixture Wishart model will be built as a variational form to realize a parametric model for PolSAR image classification. In the experiments, it will be proved that the class centers are easier to distinguish among different terrains learned from the proposed variational model. In addition, a classification performance on the PolSAR image is superior to the original point estimation Wishart model on both visual classification result and accuracy.

Polarimetric synthetic aperture radar (PolSAR) image classification is an important part of SAR data interpretation and provides more intuitive and detailed SAR polarization information. To bridge the PolSAR data and applications, it is necessary to design a comprehensive PolSAR classification framework to achieve satisfactory results. The deep neural network (DNN) appears to be a solution for the classification issue, in which it outperforms the classical supervised classifiers under the condition of sufficient training data. However, the volume of training data will greatly limit the effectiveness of practical applications. In this article, we try to solve the dependence issue on training data in three different ways: recurrent learning, data augmentation, and postprocessing. First, the long short-term memory (LSTM) network is introduced to achieve pixel sequence learning by taking into account the spatial and polarimetric features. Second, the random neighbor pixel-block (RNPB) method is proposed to increase the number of training samples for sequence learning. Third, the conditional random field (CRF) model is employed to further improve the classification accuracy. In the experiments, three sets of PolSAR data are used to evaluate the small sample performance of the proposed classification method. With only 0.5% labeled pixels for training, the proposed RNPB-LSTM-CRF method can approach 99% overall classification accuracy for all the data sets. Compared with the existing methods, the proposed method can achieve state-of-the-art results for PolSAR image classification under the condition of 1% training samples.

Although deep learning-based methods have been successfully applied to polarimetric synthetic aperture radar (PolSAR) image classification tasks, most of the available techniques are not suitable to deal with PolSAR data on irregular domains, e.g., superpixel graphs, because they are naturally designed as grid-based architectures in Euclidean space. To overcome this limitation and achieve robust PolSAR image classification, this article proposes the multiscale evolving weighted graph convolutional network, where weighted graphs based on superpixel technique and Wishart-derived distance are constructed to enable efficient handling of graphical PolSAR data representations. In this article, we derive a new architectural design named graph evolving module that combines pairwise latent feature similarity and kernel diffusion to refine the graph structure in each scale. Finally, we propose a graph integration module based on self-attention to perform robust hierarchical feature extraction and learn an optimal linear combination of various scales to exploit effective feature propagation on multiple graphs. We validate the superiority of proposed approach on classification performance with four real-measured datasets and demonstrate significant improvements compared to state-of-the-art methods. Additionally, the proposed method has shown strong generalization capacity across datasets with similar land covers.

With the improvement of spatial resolution of polarimetric synthetic aperture radar (PolSAR ) images, there was more research on texture features extracted from PolSAR images; however, the performance of texture features derived from different descriptors in PolSAR images was not fully explored. In this study, the performance of combining polarization features with texture features extracted by four descriptors, respectively, in land cover classification was explored. The experimental results indicated that: (1) different texture features have different abilities to distinguish ground objects; (2) with the improvement of spatial resolution of PolSAR images, texture features should not be ignored; (3) polarization features are very important for land cover classification, texture features can be used as auxiliary features to further improve the classification accuracy. This study could serve as a meaningful reference for information extraction from PolSAR images.

Multi-Layer Perceptrons (MLPS) are powerful function approximators. In the last decades they were successfully applied to many different regression and classification problems. Their characteristics and convergence properties are well studied and relatively well understood, but they were originally designed to work with real-valued data. The main focus of this paper is the classification of polarimetric synthetic aperture radar (POLSAR) data which are a complex-valued signal. Instead of using an arbitrarily projection of this complex-valued data to the real domain, the paper proposes the usage of complex-valued MLPS (CV-MLPS), which are an extension of MLPS to the complex domain. The paper provides a detailed yet general derivation of the complex backpropagation algorithm and mentions related problems as well as possible solutions. Furthermore, it evaluates the performance of CV-MLPS in a land-cover classification task in POLSAR images under several learning conditions, and compares the proposed classifier with standard methods. The experimental results show that CV-MLPS are successfully applicable to classification tasks in POLSAR data. They show good convergence properties and a better performance if compared to real-valued MLPS.

Convolutional neural networks (CNNs) have become the state-of-the-art in optical image processing. Recently, CNNs have been used in polarimetric synthetic aperture radar (PolSAR) image classification and obtained promising results. Unlike optical images, the unique phase information of PolSAR data expresses the structure information of objects. This special data representation makes 3D convolution which explicitly modeling the relationship between polarimetric channels perform better in the task of PolSAR image classification. However, the development of deep 3D-CNNs will cause a huge number of model parameters and expensive computational costs, which not only leads to the decrease of the interpretation speed during testing, but also greatly increases the risk of over-fitting. To alleviate this problem, a lightweight 3D-CNN framework that compresses 3D-CNNs from two aspects is proposed in this paper. Lightweight convolution operations, i.e., pseudo-3D and 3D-depthwise separable convolutions, are considered as low-latency replacements for vanilla 3D convolution. Further, fully connected layers are replaced by global average pooling to reduce the number of model parameters so as to save the memory. Under the specific classification task, the proposed methods can reduce up to 69.83% of the model parameters in convolution layers of the 3D-CNN as well as almost all the model parameters in fully connected layers, which ensures the fast PolSAR interpretation. Experiments on three PolSAR benchmark datasets, i.e., AIRSAR Flevoland, ESAR Oberpfaffenhofen, EMISAR Foulum, show that the proposed lightweight architectures can not only maintain but also slightly improve the accuracy under various criteria.

In recent years, deep learning neural networks have seen wide adoption in synthetic aperture radar (SAR) image applications. Comparatively, convenient and fast neural network models have attracted less attention. In this letter, a novel incremental Wishart broad learning system (IWBLS) is specifically designed to achieve polarimetric SAR (PolSAR) image classification for the first time. IWBLS can effectively transfer essential Wishart distribution and other types of polarimetric decomposition and spatial features to establish mapped feature and enhancement nodes in one layer without deep learning structures, which means that massive layer-by-layer training consumption can be decreased significantly. Incremental learning concept is incorporated to deal with new PolSAR images or additional features, thereby avoiding retraining entire neural networks, whose properties are very appropriate for long-term monitoring or stepwise feature integration. The experiments substantiate advantages of PolSAR image classification based on our proposed IWBLS algorithm.

Recently, convolutional neural networks (CNNs) have been successfully utilized in polarimetric synthetic aperture radar (PolSAR) image classification and obtained promising results. However, most CNN-based classification methods require a large number of labeled samples and it is difficult to obtain sufficient labeled samples. For this reason, an ensemble dual-branch CNN (EDb-CNN) is proposed for PolSAR image classification with small samples. First, to solve the problem of the small sample in PolSAR image classification, a new data enhancement method based on the superpixel algorithm is proposed to expand the number of labeled samples. Second, to obtain different scales of features from PolSAR images, a Db-CNN model is proposed. This model contains two parallel CNN structures. One CNN branch is used to extract the polarization features from the complex coherency matrix. The other branch is utilized to extract the spatial features based on weighted spatial neighborhood. On the top of these two branches, a feature fusion model is adopted to combine these two deep features, and a weighted loss function is employed to improve the learning procedure. Then, the ensemble learning algorithm is used for each CNN branch and Db-CNN network to obtain the better classification results. Finally, a postprocess algorithm based on the superpixel algorithm is proposed to improve the consistency of classification results. Experiments on two PolSAR datasets show that the proposed method achieves a much better performance than other classification methods, especially when only a few labeled samples are available.