Synthetic Aperture Radar (SAR) image segmentation is a difficult problem due to the presence of strong multiplicative noise. To attain multi-region segmentation for SAR images, this paper presents a parametric segmentation method based on the multi-texture model with level sets. Segmentation is achieved by solving level set functions obtained from minimizing the proposed energy functional. To fully utilize image information, edge feature and region information are both included in the energy functional. For the need of level set evolution, the Ratio of Exponentially Weighted Averages (ROEWA) operator is modified to obtain edge feature. Region information is obtained by the Improved Edgeworth Series Expansion (IESE), which can adaptively model a SAR image distribution with respect to various kinds of regions. The performance of the proposed method is verified by three high resolution SAR images. The experimental results demonstrate that SAR images can be segmented into multiple regions accurately without any speckle pre-processing steps by the proposed method.

The completeness of road extraction is very important for road application. However, existing deep learning (DP) methods of extraction often generate fragmented results. The prime reason is that DP-based road extraction methods use square kernel convolution, which is challenging to learn long range contextual relationships of roads. The road often produce fractures in the local interference area. Besides, the quality of extraction results will be subjected to the resolution of remote sensing (RS) image. Generally, an algorithm will produce worse fragmentation when the used data differs from the resolution of the training set. To address these issues, we propose a novel road extraction framework for RS images, named the Multi-Scale and Multi-Direction Feature Fusion Network (MSMDFF-Net). This framework comprises three main components: the Multi-Directional Feature Fusion (MDFF) Initial Block, the Multi-Scale Residual (MSR) encoder, and the Multi-Directional Combined Fusion (MDCF) decoder. Firstly, according to the road’s morphological characteristics, we develop a strip convolution module with a direction parameter (SCM-D). Then, to make the extracted result more complete, four SCM-D with different directions are used to MDFF-Initial Block and MDCF-decoder. Finally, we incorporate an additional branch into the ResNet encoding module to build MSR-encoder for improving the generalization of the model on different resolution RS image. Extensive experiments on three popular datasets with different resolution (Massachusetts, DeepGlobe, and SpaceNet datasets) show that the proposed MSMDFF-Net achieves new state-of-the-art results. The code will be available at https://github.com/wycloveinfall/MSMDFF-NET.

Roads are an important recognition target in synthetic aperture radar (SAR) image interpretation. Although a considerable number of high-quality SAR images are now available, the method of road extraction is lagging. To extract the road network with low missed and false rates, this paper proposed a road network extraction approach which includes line detection, road segmentation, road network extraction and optimization. First, the linear feature response and direction map are obtained from the SAR intensity image using the multiplicative Duda operation. Then, the backscattering coefficient and coefficient of variation are combined using a support vector machine to eliminate the linear structures of non-roads, and the binary image of road candidates is subsequently achieved by morphological profiles of path openings. Next, with the obtained direction map, a novel thinning method based on binary image decomposition and curve fitting is presented to obtain line segments of the road network. Finally, a series of measures which involve overlap, continuity, and junction optimization are proposed to construct the road network. In the experiments, the proposed method was applied to Radarsat-2 and TerraSAR-X high-resolution images. The experimental results showed that the proposed method had an excellent performance in terms of both completeness and correctness.

In this article, a method for landslides identification based on polarimetric decomposition techniques and the Wishart classifier is presented. Several parameters of Cloude decomposition and Freeman–Durden decomposition are introduced to analyse the radar scattering mechanisms over landslides and their surroundings. By analysing the polarimetric characteristics of three typical landslides triggered by the 2008 Wenchuan earthquake, the result demonstrates that the dominant scattering mechanism of the landslides is the surface scattering component, and the scattering entropy is centred between 0.6 and 0.8. Besides, the result also indicates that the surroundings show the dominance of the volume scattering component, whereas their scattering entropy is almost similar to the landslides. According to the experimental analysis, two temporal SAR images are classified into nine categories, respectively, based on polarimetric decomposition techniques and the Wishart classifier. By comparing the surface scattering areas in the second image with the volume scattering areas in the first image, and combining the scattering entropy, the new slope failures are obtained using the change detection method. In addition, in order to reduce the errors introduced by the registration and noises, the identification map of new landslides is processed by the morphology algorithm. Finally, a field experiment is carried out to verify the existence of the new slope failures and thereby the result shows the validity of the proposed method.

Low-grade roads have complex features such as geometry, reflection spectrum, and spatial topology in remotely sensing optical images due to the different materials of those roads and also because they are easily obscured by vegetation or buildings, which leads to the low accuracy of low-grade road extraction from remote sensing images. To address this problem, this paper proposes a novel deep learning network referred to as SDG-DenseNet as well as a fusion method of optical and Synthetic Aperture Radar (SAR) data on decision level to extract low-grade roads. On one hand, in order to enlarge the receptive field and ensemble multi-scale features in commonly used deep learning networks, we develop SDG-DenseNet in terms of three modules: stem block, D-Dense block, and GIRM module, in which the Stem block applies two consecutive small-sized convolution kernels instead of the large-sized convolution kernel, the D-Dense block applies three consecutive dilated convolutions after the initial Dense block, and Global Information Recovery Module (GIRM) combines the ideas of dilated convolution and attention mechanism. On the other hand, considering the penetrating capacity and oblique observation of SAR, which can obtain information from those low-grade roads obscured by vegetation or buildings in optical images, we integrate the extracted road result from SAR images into that from optical images at decision level to enhance the extraction accuracy. The experimental result shows that the proposed SDG-DenseNet attains higher IoU and F1 scores than other network models applied to low-grade road extraction from optical images. Furthermore, it verifies that the decision-level fusion of road binary maps from SAR and optical images can further significantly improve the F1, COR, and COM scores.

Multi-object semantic segmentation from remote sensing images has gained significant attention in land resource surveying, global change monitoring, and disaster detection. Compared to other application scenarios, the objects in the remote sensing field are larger and have a wider range of distribution. In addition, some similar targets, such as roads and concrete-roofed buildings, are easily misjudged. However, existing convolutional neural networks operate only in the local receptive field, and this limits their capacity to represent the potential association between different objects and surrounding features. This paper develops a Multi-task Quadruple Attention Network (MQANet) to address the above-mentioned issues and increase segmentation accuracy. The MQANet contains four attention modules: position attention module (PAM), channel attention module (CAM), label attention module (LAM), and edge attention module (EAM). The quadruple attention modules obtain global features by expanding the receptive fields of the network and introducing spatial context information in the label. Then, a multi-tasking mechanism which splits a multi-category segmentation task into several binary-classification segmentation tasks is introduced to improve the ability to identify similar objects. The proposed MQANet network was applied to the Potsdam dataset, the Vaihingen dataset and self-annotated images from Chongzhou and Wuzhen (CZ-WZ), representative cities in China. Our MQANet performs better over the baseline net by a large margin of +6.33 OA and +7.05 Mean F1-score on the Vaihingen dataset, +3.57 OA and +2.83 Mean F1-score on the Potsdam dataset, and +3.88 OA and +8.65 Mean F1-score on the self-annotated dataset (CZ-WZ dataset). In addition, each image execution time of the MQANet model is reduced 66.6 ms compared to UNet. Moreover, the effectiveness of MQANet was also proven by comparative experiments with other studies.