Synthetic aperture radar (SAR) has played an important role in many fields. The breakthroughs in deep learning methods represented by the convolutional neural network (CNN) models have greatly improved the SAR image recognition accuracy. Classification models based on CNN can perform high-precision classification, but there are security problems with adversarial examples (AEs). However, the research on AEs is mostly limited to nature image, and remote sensing images (SAR, multispectral, etc.) have not been extensively studied. To explore the basic characteristics of adversarial examples of SAR images (ASIs), we use two classic white-box attack methods to generate ASIs from two SAR image classification datasets and then evaluate the vulnerability of six commonly used CNNs. The results show that ASIs are quite effective in fooling CNNs trained on SAR images as testified by the obtained high attack success rate. Due to the structural differences among CNNs, different CNNs present different vulnerabilities in the face of ASIs. We found that ASIs generated by nontarget attack algorithms feature attack selectivity, which is related to the feature space distribution of the clean SAR images and the decision boundary of the classification model. We proposed the sample-boundary-based adversarial example selectivity distance (AESD) to successfully explain the attack selectivity of ASIs, which provided the theoretical basis for targeted attacks. We also analyzed the effects of image parameters such as image size and the number of channels on the attack success rate of ASIs through parameter sensitivity, providing detailed evidence for the design of ASIs defense methods.

Landslide hazard assessment is essential for determining the probability of landslide occurrence in a specific spatial and temporal range. The hazard assessment of potential landslides could support landslide disaster early warning and disaster prevention decisions, which have important guiding significance for urban construction and sustainable development. Due to the lack of consideration of the synergistic effect of multiple factors and geographic scene heterogeneity, the accuracy of existing landslide hazard assessment methods still needs to be improved, and the interpretability and applicability of existing models still need to be improved. In this paper, we propose a landslide hazard assessment method considering the synergistic effect of multiple factors, including natural factors and human activities, and the heterogeneity of geographic scenes. On this basis, we carry out experimental verification on rainfall–induced landslides in Dehong Prefecture, Yunnan Province, China. Firstly, rainfall–induced landslide hazards’ characteristics and impact factors are analyzed and classified. The whole study area is divided into some homogeneous sub–regions using regional dynamic constraint clustering based on the similarity of underlying environmental variables. Then, considering the spatial autocorrelation between various landslide conditioning and trigger factors, a local weighted random forest model is developed to evaluate the rainfall–induced landslide hazards comprehensively. Experimental results show that the proposed method has higher accuracy and interpretability than the existing representative methods and can provide useful references for preventing landslide hazards.

Understanding the spatial distribution and associated factors of sustainable environmental development (SED) is crucial for effectively formulating action plans aimed at achieving the environment-related Sustainable Development Goals (SDGs). Because of spatial heterogeneity characterized by non-uniform distributions of SED within a region, the use of a single or aggregated value at the national or provincial scale in existing research obscures internal variabilities and fails to adequately reveal the spatial disparities in the progress of environment-related SDGs. Consequently, this study aims to investigate the spatially heterogeneous patterns of SED in prefecture-level cities within the Yangtze River Economic Belt (YREB), China—a critical economic zone characterized by stark intra-regional disparities and pressing environmental challenges. To achieve this objective, the SED index is first constructed using principal component weighted aggregation to quantify the SED status, local Moran’s I is then employed to identify heterogenous patters of the spatial distribution patterns of SED, and geographical random forest is utilized to explore spatially varying association patterns between SED and influences factors. In the YREB, the findings indicate the following key insights: (1) SED has generally shown a positive trend across most cities from 2013 to 2021, with approximately 30% experiencing a downward trend, particularly in Jiangxi, Hubei, and Hunan Provinces; (2) the number of high- or low-value aggregation clusters has decreased, concurrent with an increase in spatial variability; (3) human activity disturbance and population density have been identified as significant factors influencing SED, with a notable impact in cities across Sichuan, Chongqing, Guizhou, and Hubei Provinces. This research contributes technical support and a scientific foundation for evaluating and enhancing SED.

Accurately and efficiently extracting building footprints from a wide range of remote sensed imagery remains a challenge due to their complex structure, variety of scales and diverse appearances. Existing convolutional neural network (CNN)-based building extraction methods are complained that they cannot detect the tiny buildings because the spatial information of CNN feature maps are lost during repeated pooling operations of the CNN, and the large buildings still have inaccurate segmentation edges. Moreover, features extracted by a CNN are always partial which restricted by the size of the respective field, and large-scale buildings with low texture are always discontinuous and holey when extracted. This paper proposes a novel multi attending path neural network (MAP-Net) for accurately extracting multiscale building footprints and precise boundaries. MAP-Net learns spatial localization-preserved multiscale features through a multi-parallel path in which each stage is gradually generated to extract high-level semantic features with fixed resolution. Then, an attention module adaptively squeezes channel-wise features from each path for optimization, and a pyramid spatial pooling module captures global dependency for refining discontinuous building footprints. Experimental results show that MAP-Net outperforms state-of-the-art (SOTA) algorithms in boundary localization accuracy as well as continuity of large buildings. Specifically, our method achieved 0.68\%, 1.74\%, 1.46\% precision, and 1.50\%, 1.53\%, 0.82\% IoU score improvement without increasing computational complexity compared with the latest HRNetv2 on the Urban 3D, Deep Globe and WHU datasets, respectively. The TensorFlow implementation is available at https://github.com/lehaifeng/MAPNet.