Article
Environmental Sciences
Liegang Xia, Xiongbo Zhang, Junxia Zhang, Haiping Yang, Tingting Chen
Summary: This study introduces a semi-supervised deep learning approach based on an edge detection network for extracting building roof boundaries from high-resolution remote sensing images, achieving satisfactory results with a small number of labeled samples and abundant unlabeled images used for joint training.
Article
Environmental Sciences
Khaled Moghalles, Heng-Chao Li, Zaid Al-Huda, Essa Abdullah
Summary: This paper proposes a framework that utilizes deep seeds and optimal segmentation to extract buildings from high-resolution satellite images. By inputting cropped images instead of resized images to a deep convolutional neural network, accurate semantic segmentation of buildings is achieved. By locating deep seeds in the images and predicting boundary maps, and through the construction of a hierarchical segmentation tree and the development of a graphical model, the final optimized building segmentation is achieved.
JOURNAL OF APPLIED REMOTE SENSING
(2022)
Article
Engineering, Electrical & Electronic
Jianfeng Huang, Xinchang Zhang, Ying Sun, Qinchuan Xin
Summary: This article introduces an attention-guided label refinement network (ALRNet) for improved semantic labeling of VHR images, which progressively refines the coarse labeling maps of different scales using a channelwise attention mechanism to bridge the semantic gap between features of different levels.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2021)
Article
Automation & Control Systems
Sihan Yang, Qiang He, Jae Hak Lim, Gwanggil Jeon
Summary: Automatic building extraction is a significant research topic in the field of remote sensing. The proposed boundary-guided DCNN method improves the accuracy of building extraction by integrating boundary information and mask features. Experimental results demonstrate that this method outperforms state-of-the-art models on multiple datasets.
INTERNATIONAL JOURNAL OF ADVANCED MANUFACTURING TECHNOLOGY
(2022)
Review
Geography, Physical
Jiayi Li, Xin Huang, Lilin Tu, Tao Zhang, Leiguang Wang
Summary: This article provides a comprehensive review of the recent advancements in building extraction from very high resolution (VHR) optical remote sensing images. It categorizes the building detection methods into physical rule based methods, image segmentation based methods, and traditional and advanced machine learning methods. The article also discusses four promising research directions in building extraction and provides a better understanding of this topic for researchers.
GISCIENCE & REMOTE SENSING
(2022)
Article
Engineering, Electrical & Electronic
Yanjing Lei, Jiamin Yu, Sixian Chan, Wei Wu, Xiaoying Liu
Summary: Building extraction is crucial in high-resolution remote sensing image processing, serving as the foundation for urban planning and demographic analysis. This paper presents a novel deep learning network called SNLRUX++ for building extraction. Experimental results demonstrate the effectiveness and generalization ability of the proposed method.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2022)
Article
Chemistry, Analytical
Jia Song, A-Xing Zhu, Yunqiang Zhu
Summary: This article explores the role of Vision Transformer networks in extracting building footprints from high-resolution satellite images. Different hyperparameter values were used to design and compare Transformer-based models, and their impact on accuracy was analyzed. The results suggest that smaller image patches and higher-dimensional embeddings contribute to higher accuracy. Furthermore, the Transformer-based network is shown to be scalable and can be trained with general-scale GPUs while achieving higher accuracy than convolutional neural networks.
Article
Environmental Sciences
Dehua Xie, Han Xu, Xiliu Xiong, Min Liu, Haoran Hu, Mengsen Xiong, Luo Liu
Summary: Accurate cropland information is crucial for food security assessment and agricultural policy formulation. However, extracting cropland from VHR remote sensing images in southern China is challenging due to heterogeneity and limited observations. To address this, we proposed a deep learning-based method using an improved HRRS-U-Net model.
Article
Computer Science, Artificial Intelligence
Maoguo Gong, Tongfei Liu, Mingyang Zhang, Qingfu Zhang, Di Lu, Hanhong Zheng, Fenlong Jiang
Summary: This paper proposes a context-content collaborative network (C3Net) with an encoder-decoder structure to achieve a good trade-off between precision and completeness in building extraction. The C3Net consists of a context-content aware module (C2AM) and an edge residual refinement module (ER2M) that capture the localization information of buildings and refine the features of decoder output, respectively. Extensive experiments show that the C3Net achieves competitive performance and a better trade-off between precision and completeness.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Environmental Sciences
Liang Huang, Qiuyuan Tian, Bo-Hui Tang, Weipeng Le, Min Wang, Xianguang Ma
Summary: A change detection network for building VHR remote sensing images based on Siamese EfficientNet B4-MANet (Siam-EMNet) is proposed, which integrates dual attention mechanism to accurately detect change regions and rough edges of buildings.
Article
Geochemistry & Geophysics
Xiaokang Zhang, Boning Zhang, Weikang Yu, Xudong Kang
Summary: Deep convolutional neural networks (DCNNs) are widely used in object extraction from very-high-resolution (VHR) remote sensing images. However, the scarcity of labeled local datasets affects the prediction performances of DCNNs, and privacy concerns arise in traditional deep learning schemes. To address these issues, a novel federated learning scheme with prototype matching (FedPM) is proposed, which collaboratively learns a richer DCNN model by leveraging distributed remote sensing data. This scheme ensures data privacy by conducting federated optimization in the gradient space. Experimental results on VHR aerial and satellite image datasets demonstrate that FedPM significantly improves the prediction performance of DCNNs with lower communication costs. This is the first application of federated learning in remote sensing visual tasks.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Haonan Guo, Qian Shi, Bo Du, Liangpei Zhang, Dongzhi Wang, Huaxiang Ding
Summary: This study proposes a new method for building extraction using a multitask parallel attention convolutional network to recognize buildings under different scenes, and integrates the results with a simple postprocessing method, achieving better performance compared to existing algorithms.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2021)
Article
Computer Science, Artificial Intelligence
Hongyan Zhang, Yue Liao, Honghai Yang, Guangyi Yang, Liangpei Zhang
Summary: In this article, a novel local-global dual-stream network (DS-Net) is proposed for accurate mapping of building rooftops in very-high-resolution RS images. The DS-Net captures local and long-range information adaptively and works in a complementary manner with different fields of view. The proposed model achieves competitive or superior performance compared to current state-of-the-art methods on three widely used datasets.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Geography, Physical
Wufan Zhao, Claudio Persello, Alfred Stein
Summary: Roof structure information is crucial for creating detailed 3D building models. This paper introduces a fast and parsimonious method to extract building rooflines and structures from high resolution remote sensing imagery. The proposed method outperforms competing models in both qualitative and quantitative evaluations.
ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING
(2022)
Article
Computer Science, Information Systems
Ramesh Raghavan, Dinesh Chander Verma, Digvijay Pandey, Rohit Anand, Binay Kumar Pandey, Harinder Singh
Summary: Building extraction is crucial in urban dynamics for disaster management, change detection, and population estimation. However, extracting buildings from satellite data is challenging due to variations in illumination and structure. To overcome this, a convolutional neural network and Mask-RCNN algorithm with advanced image augmentation technique are proposed. The results show improved accuracy in building extraction.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Review
Environmental Sciences
Mingqiang Guo, Zhongyang Yu, Yongyang Xu, Ying Huang, Chunfeng Li
Summary: Mangroves play a crucial role in ecosystem services. A pixel classification model inspired by deep learning technology was proposed for accurately extracting mangroves. Experimental results show that the trained mangrove extraction network achieves an overall accuracy of 97.48%.
Article
Environmental Sciences
Jie Wan, Zhong Xie, Yongyang Xu, Ziyin Zeng, Ding Yuan, Qinjun Qiu
Summary: The DGANet is a Dilated Graph Attention-based Network designed to extract local geometric features by establishing local dilated graph-like regions and integrating dilated graph attention modules.
Article
Environmental Sciences
Anna Hu, Siqiong Chen, Liang Wu, Zhong Xie, Qinjun Qiu, Yongyang Xu
Summary: An improved generative adversarial network, named WSGAN, was designed in this study to extract road networks through a weakly supervised process. The method involves generating mapping images and post-processing binary images, achieving high accuracy scores in road network extraction from remote sensing images.
Article
Astronomy & Astrophysics
Huihui Cai, Siqiong Chen, Yongyang Xu, Zixuan Li, Xiangjin Ran, Xingping Wen, Yongsheng Li, Yanqing Men
Summary: Novel prediction methods using artificial intelligence have been developed to improve the identification, discovery, and utilization of new types of mineral resources, but the lack of large training data sets remains a challenge. To address this, a semi-supervised machine-learning method was developed to identify metallogenic anomalies, showing irregular distributions matching known mineralization areas. This method's accuracy was confirmed through interdisciplinary intelligent analysis, indicating its potential for improving regional geological surveys.
EARTH AND SPACE SCIENCE
(2021)
Article
Environmental Sciences
Yongyang Xu, Wei Luo, Anna Hu, Zhong Xie, Xuejing Xie, Liufeng Tao
Summary: This study proposes an improved generative adversarial network with self-attention and texture enhancement (TE-SAGAN) for remote sensing super-resolution images. The proposed method successfully addresses issues such as blurry object edges and existing artifacts through the use of self-attention mechanism, texture enhancement, and a joint loss function.
Article
Geography
Yongyang Xu, Zhanjun He, Xuejing Xie, Zhong Xie, Jing Luo, Hong Xie
Summary: A learning strategy based on multiple features and context information is developed to detect the function of single buildings in a vector map. Experimental results show that the proposed method can learn local and contextual building information and outperforms other machine learning methods in building function classification.
TRANSACTIONS IN GIS
(2022)
Article
Environmental Sciences
Ziyin Zeng, Yongyang Xu, Zhong Xie, Jie Wan, Weichao Wu, Wenxia Dai
Summary: This paper proposes a random graph-based graph convolution network, RG-GCN, to address the issue of insufficient samples in point cloud semantic segmentation. Through data augmentation and feature extraction, the network achieves excellent performance on indoor and outdoor datasets.
Article
Chemistry, Multidisciplinary
Xuejing Xie, Yawen Liu, Yongyang Xu, Zhanjun He, Xueye Chen, Xiaoyun Zheng, Zhong Xie
Summary: The functional classification of buildings is important for urban planning and government departments. A semi-supervised graph structure network combined with a unified message passing model is introduced for building function recognition. By utilizing the spatial distribution, characteristics, and POIs information of buildings, this method can capture more meaningful information with limited labels and achieve better classification results.
APPLIED SCIENCES-BASEL
(2022)
Article
Astronomy & Astrophysics
Bin Feng, Lirong Chen, Yongyang Xu, Yu Zhang
Summary: Based on stream sediment data in the Cu-Zn-Ag metallogenic area in southwest Fujian province, this study used three deep learning models: autoencoder (AE), multi-convolutional autoencoder (MCAE), and fusion convolutional autoencoder (FCAE) to extract features for geochemical anomaly identification. The results showed that FCAE had the highest consistency with known copper mineral occurrences, followed by MCAE and AE. FCAE combined the advantages of MCAE and AE, focusing more on structural distribution and mixed features.
EARTH AND SPACE SCIENCE
(2022)
Article
Environmental Sciences
Liufeng Tao, Yuqiong Cui, Yongyang Xu, Zhanlong Chen, Han Guo, Bo Huang, Zhong Xie
Summary: Urban fires pose threats to the economy and safety of urban residents. An optimized placement of fire stations is necessary to cover as many areas as possible and focus on high-risk areas. This study proposes a multi-objective optimization model for fire station planning based on the backup coverage model, using the SAVEE model to quantify the spatial heterogeneity of urban fires. The results demonstrate the effectiveness of the proposed model in covering high-risk areas with as few fire stations as possible.
INTERNATIONAL JOURNAL OF ENVIRONMENTAL RESEARCH AND PUBLIC HEALTH
(2023)
Article
Geochemistry & Geophysics
Anna Hu, Liang Wu, Siqiong Chen, Yongyang Xu, Haitao Wang, Zhong Xie
Summary: This article proposes a building shape-preserving framework to solve the imperfect boundary problem and eliminate sawtooth noise from building extractions. By using instance segmentation method and boundary network, the building boundaries can be extracted more accurately, and a footprint information evaluation algorithm is used to evaluate the extracted building shape. The experimental results demonstrate the effectiveness of the proposed method.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Remote Sensing
Yongyang Xu, Wei Tang, Ziyin Zeng, Weichao Wu, Jie Wan, Han Guo, Zhong Xie
Summary: 3D point cloud semantic segmentation is crucial for understanding 3D environments. Existing approaches of local context learning in point clouds are based on predefined neighbors, but K-nearest neighbor algorithm (KNN) is suboptimal. This study proposes NeiEA-Net, a simple and effective network that optimizes local neighbors in high-dimensional feature space for semantic segmentation of point clouds. The network further reduces redundant information by adaptively aggregating features of different scales. Experimental results on three large-scale benchmarks demonstrate the superiority of this network.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2023)
Article
Engineering, Electrical & Electronic
Jie Wan, Ziyin Zeng, Qinjun Qiu, Zhong Xie, Yongyang Xu
Summary: This paper proposes an innovative network called PointNest, which achieves accurate point segmentation by learning multiscale point feature propagation. The introduction of a deep supervision strategy further improves training efficiency and performance. PointNest outperforms existing mainstream methods on three public benchmarks.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Remote Sensing
Ziyin Zeng, Yongyang Xu, Zhong Xie, Wei Tang, Jie Wan, Weichao Wu
Summary: In this study, a network called LEARD-Net is proposed for semantic segmentation of large-scale point cloud scene data. The network utilizes color information and employs local feature extraction and aggregation modules to effectively process the point cloud data.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2022)
Article
Engineering, Electrical & Electronic
Jie Wan, Zhong Xie, Yongyang Xu, Siqiong Chen, Qinjun Qiu
Summary: The DA-RoadNet is a road extraction network with dual attention mechanism, designed to effectively solve discontinuous problems and preserve the integrity of extracted roads by utilizing a deep learning network model and a novel attention mechanism module. Additionally, a hybrid loss function is employed to address class imbalance, ensuring stable training of the network model and avoiding local optima.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2021)