A5060209705- Remote Sensing and LiDAR Applications
- Remote-Sensing Image Classification
- Remote Sensing and Land Use
- Image Retrieval and Classification Techniques
- Remote Sensing in Agriculture
- Face and Expression Recognition
- Advanced Image and Video Retrieval Techniques
- Cryospheric studies and observations
- Automated Road and Building Extraction
- Text and Document Classification Technologies
- 3D Surveying and Cultural Heritage
- Advanced Algorithms and Applications
- Video Surveillance and Tracking Methods
- Medical Image Segmentation Techniques
- Advanced Neural Network Applications
- Landslides and related hazards
- Anomaly Detection Techniques and Applications
- Neural Networks and Applications
- Machine Learning and ELM
- Impact of Light on Environment and Health
- Satellite Image Processing and Photogrammetry
- Atmospheric aerosols and clouds
- Arctic and Antarctic ice dynamics
- Fault Detection and Control Systems
- Forest ecology and management
State Key Laboratory of Remote Sensing Science
2020-2024
Beijing Normal University
2017-2024
Shenzhen Bao'an District People's Hospital
2024
Zhejiang University of Technology
2024
Tongji University
2019-2023
China Agricultural University
2023
City University College of Science and Technology
2021
Institute of Remote Sensing and Digital Earth
2020
Sichuan Normal University
2019
Air Force Engineering University
2019
Satellite mapping of buildings and built-up areas used to be delineated from high spatial resolution (e.g., meters or sub-meters) middle tens hundreds meters) satellite images, respectively. To the best our knowledge, it is important explore a deep-learning approach delineate high-resolution semantic maps middle-resolution images. The termed as super-resolution segmentation in this paper. Specifically, we design neural network with integrated low-level image features high-level...
It is of great significance for emergency rescue to rapidly assess damage buildings after an earthquake. Some previous methods are time-consuming, data difficult obtain, or there lack regional assessment. We proposed a novel way building by comprehensively utilizing earth observation-derived and field investigation alleviate the above problems. These related hazard-causing factors, hazard-formative environment, hazard-affected body. Specifically, predicted ground motion parameters used...
Spaceborne laser altimetry technology assists global DEMs to improve the accuracy of elevation data due its highly accurate range and wide coverage. As compared previous altimeter systems used for Earth observation, ICESat-2 has a sensitivity photon detection that can provide more denser surface observations. This paper proposed DEM correction model using data. The altimetric verify errors in coverage areas firstly. Then an attribute set was constructed evaluate error sources global-scale...
Due to its high ranging accuracy, spaceborne laser altimetry technology can improve the accuracy of satellite stereo mapping without ground control points. In past, full-waveform ICE , CLOUD and Land Elevation Satellite ( ICESat ) altimeter data have been used as one main sources for global elevation control. As a second-generation satellite, ICESat-2 is equipped with an using photon counting mode. This further application capability because six beams along-track repetition frequency, which...
Rapid intelligent detection of airports from remote sensing images is required to accomplish autonomous landing unmanned aerial vehicles (UAVs) and other tasks. To address the insufficiency traditional models in detecting under complicated backgrounds images, we propose an end-to-end airport hierarchical expression model based on deep transferable convolutional neural networks. Based transfer learning, solve fundamental problem overfitting due inadequate number labeled by transferring...
Modern convolutional neural networks (CNNs) are often trained on pre-set data sets with a fixed size. As for the large-scale applications of satellite images, example, global or regional mappings, these images collected incrementally by multiple stages in general. In other words, sizes training datasets might be increased tasks mapping rather than beforehand. this paper, we present novel algorithm, called GeoBoost, incremental-learning semantic segmentation via networks. Specifically,...
During disaster response, it is very important to obtain the information of collapsed building distribution accurately and quickly. However, limited by some practical factors, existed methods often suffer from contradiction between accuracy efficiency damage extraction. This paper proposed a simple effective framework rapid recognize objects using pre-disaster maps post-disaster quasi-panchromatic remote sensing images. The method validated several historical disasters in xBD dataset tested...
To achieve high scene classification performance of spatial resolution remote sensing images (HSR-RSIs), it is important to learn a discriminative space in which the distance metric can precisely measure both similarity and dissimilarity features labels between images. While traditional learning methods focus on preserving interclass separability, label consistency (LC) less involved, this might degrade accuracy. Aiming at considering intraclass compactness HSR-RSIs, we propose method with...
As the accuracy of satellite laser altimetry is susceptible to real-time atmospheric conditions, with-in footprint topography fluctuation, and detector noise, etc., we proposed a method by comprehensively analyzing ranging error evaluation labels extract high-accuracy elevation control points that suitable for imagery based topographic mapping applications. Using ICESat data, global high dataset including more than 60 million points, on model waveform quality analysis paper. For land areas,...
In satellite laser altimetry, it is a challenging task to accurately extract peak positions from full waveforms due the overlapped or weak peaks within large footprints, which substantially affects subsequent applications. this article, improve ranging resolution and accuracy, we propose novel approach by combining deconvolution with Gaussian decomposition. The applied in two main phases: 1) first used remove system contribution (the transmit pulse spreading over several nanoseconds, noise);...
Stereo photogrammetric survey used to be extract the height of buildings, then convert number stories through certain rules estimate buildings by means satellite remote sensing. In contrast, we propose a new method using deep learning from monocular optical image end in this paper. To best our knowledge, is first attempt directly images. Specifically, proposed method, extend classic object detection network, i.e., Mask R-CNN, adding head predict detected GF-2 images nine cities China are...
Post-classification comparison using pre- and post-event remote-sensing images is a common way to quickly assess the impacts of natural disaster on buildings. Both effectiveness efficiency post-classification heavily depend classifier’s precision generalization abilities. In practice, practitioners used train novel image classifier for an unexpected from scratch in order evaluate building damage. Recently, it has become feasible deep learning model recognize buildings very high-resolution...
In this paper, we present a novel approach for automatically detecting buildings from multiple heterogeneous and uncalibrated very high-resolution (VHR) satellite images rapid response to natural disasters. the proposed method, simple efficient visual attention method is first used extract built-up area candidates (BACs) each multispectral (MS) image. After this, morphological building indices (MBIs) are extracted all masked panchromatic (PAN) MS with BACs characterize structural features of...
Stochastic gradient descent and other adaptive optimization methods have been proved effective for training deep neural networks. Within each epoch of these methods, the whole set is involved to train model. In general, large data sets redundancy among their samples. this paper, we present an algorithm reduce time consumption CNN by dropping certain samples out, which called greedy DropSample. For absence samples, distribution networks' activations biased during training. By correcting mean...
High-resolution remote-sensing imagery has proven useful for building extraction. Unfortunately, due to the high acquisition costs and infrequent availability of high-resolution imagery, low-resolution images are more practical large-scale mapping or change tracking buildings. However, extracting buildings from is a challenging task. Compared with images, pose two critical challenges in terms segmentation: effects fuzzy boundary details on lack local textures. In this study, we propose...
The building damage caused by natural disasters seriously threatens human security. Applying deep learning algorithms to identify collapsed buildings from remote sensing images is crucial for rapid post-disaster emergency response. However, the diversity of buildings, limited training dataset size, and lack ground-truth samples after sudden can significantly reduce generalization a pre-trained model identification when applied directly non-preset locations. To address this challenge,...
Field survey is a labour-intensive way to objectively evaluate the grade of building damage triggered by earthquakes. In this paper, we present decision-tree-based approach classify type using multiple-source remote sensing from both pre- and postearthquakes. Specifically, boundary buildings delineated preearthquake satellite images an unsupervised learning method. Then, classified into four types decision tree method postearthquake UAV images, that is, basically intact buildings, slightly...