A5000877601- Remote-Sensing Image Classification
- Remote Sensing and Land Use
- Remote Sensing and LiDAR Applications
- Automated Road and Building Extraction
- Advanced Image and Video Retrieval Techniques
- Remote Sensing in Agriculture
- Geographic Information Systems Studies
- Advanced Neural Network Applications
- Geology and Paleoclimatology Research
- Land Use and Ecosystem Services
- Pleistocene-Era Hominins and Archaeology
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Advanced Image Fusion Techniques
- Time Series Analysis and Forecasting
- Tree-ring climate responses
- Robotics and Sensor-Based Localization
- Geochemistry and Geologic Mapping
- 3D Surveying and Cultural Heritage
- Video Surveillance and Tracking Methods
- Archaeology and ancient environmental studies
- Wildlife-Road Interactions and Conservation
- Data-Driven Disease Surveillance
- Isotope Analysis in Ecology
- Infrastructure Maintenance and Monitoring
Stanford University
2023-2025
State Key Laboratory of Information Engineering in Surveying Mapping and Remote Sensing
2018-2024
Wuhan University
2018-2024
The University of Texas Health Science Center at Houston
2024
Sichuan University
2024
West China Hospital of Sichuan University
2024
Guangxi Science and Technology Department
2024
Sun Yat-sen University
2011-2023
Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou)
2022-2023
Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai)
2022-2023
Geospatial object segmentation, as a particular semantic segmentation task, always faces with larger-scale variation, larger intra-class variance of background, and foreground-background imbalance in the high spatial resolution (HSR) remote sensing imagery. However, general methods mainly focus on scale variation natural scene, inadequate consideration other two problems that usually happen large area earth observation scene. In this paper, we argue lie lack foreground modeling propose...
Road detection and centerline extraction from very high-resolution (VHR) remote sensing imagery are of great significance in various practical applications. operations depend on each other, to a certain extent. The road constrains the appearance centerline, enhances linear features detection. However, most previous works have addressed these two tasks separately not considered symbiotic relationship between them, making it difficult obtain smooth complete roads. In this paper, novel...
Deep learning approaches have shown promising results in remote sensing high spatial resolution (HSR) land-cover mapping. However, urban and rural scenes can show completely different geographical landscapes, the inadequate generalizability of these algorithms hinders city-level or national-level Most existing HSR datasets mainly promote research semantic representation, thereby ignoring model transferability. In this paper, we introduce Land-cOVEr Domain Adaptive segmentation (LoveDA)...
Deep learning techniques have been widely applied to hyperspectral image (HSI) classification and achieved great success. However, the deep neural network model has a large parameter space requires number of labeled data. methods for HSI usually follow patchwise framework. Recently, fast patch-free global (FPGA) architecture was proposed according spatial context information. FPGA difficulty in extracting most discriminative features when sample data are imbalanced. In this article,...
The small object semantic segmentation task is aimed at automatically extracting key objects from high-resolution remote sensing (HRS) imagery. Compared with the large-scale coverage areas for imagery, objects, such as cars and ships, in HRS imagery often contain only a few pixels. In this article, to tackle problem, foreground activation (FA)-driven (FactSeg) framework proposed perspectives of structure optimization. design, FA representation enhance awareness weak features objects. made up...
Deep learning algorithms, especially convolutional neural networks (CNNs), have recently emerged as a dominant paradigm for high spatial resolution remote sensing (HRS) image recognition. A large amount of CNNs already been successfully applied to various HRS recognition tasks, such land-cover classification and scene classification. However, they are often modifications the existing derived from natural processing, in which network architecture is inherited without consideration complexity...
For high spatial resolution (HSR) remote sensing images, bitemporal supervised learning always dominates change detection using many pairwise labeled images. However, it is very expensive and time-consuming to label large-scale HSR In this paper, we propose single-temporal (STAR) for from a new perspective of exploiting object changes in unpaired images as supervisory signals. STAR enables us train high-accuracy detector only generalize real-world To evaluate the effectiveness STAR, design...
Geospatial object segmentation, a fundamental Earth vision task, always suffers from scale variation, the larger intra-class variance of background, and foreground-background imbalance in high spatial resolution (HSR) remote sensing imagery. Generic semantic segmentation methods mainly focus on variation natural scenarios. However, other two problems are insufficiently considered large area observation In this paper, we propose foreground-aware relation network (FarSeg++) perspectives...
Abstract. Here we describe LegacyClimate 1.0, a dataset of the reconstruction mean July temperature (TJuly), annual (Tann), and precipitation (Pann) from 2594 fossil pollen records Northern Hemisphere, spanning entire Holocene, with some reaching back to Last Glacial Period. Two methods, modern analog technique (MAT) weighted averaging partial least squares regression (WA-PLS), reveal similar results regarding spatial temporal patterns. To reduce impact on reconstruction, vice versa, also...
Abstract. A mismatch between model- and proxy-based Holocene climate change, known as the “Holocene conundrum”, may partially originate from poor spatial coverage of reconstructions in, for example, Asia, limiting number grid cells model–data comparisons. Here we investigate hemispheric, latitudinal, regional mean time series time-slice anomaly maps pollen-based annual temperature, July precipitation 1908 records in Northern Hemisphere extratropics. Temperature trends show strong latitudinal...
Our understanding of the temporal dynamics Earth's surface has been significantly advanced by deep vision models, which often require a massive amount labeled multi-temporal images for training. However, collecting, preprocessing, and annotating remote sensing at scale is non-trivial since it expensive knowledge-intensive. In this paper, we present scalable change data generators based on generative are cheap automatic, alleviating these problems. main idea to simulate stochastic process...
Remote sensing image scene classification is a challenging task. With the development of deep learning, methods based on convolutional neural networks (CNNs) have made great achievements in remote classification. Since training CNN requires large number labeled samples, generative adversarial network (GAN) for sample generation represents new opportunity to solve problem limited samples. However, most existing GAN-based can only generate unlabeled instead samples with corresponding category....
Road extraction from very high-resolution (VHR) remote sensing imagery remains a huge challenge, due to the shadows and occlusions of trees buildings. Such complex backgrounds result in deep networks often producing fragmented roads with poor connectivity. has three typical tasks: road surface segmentation (SS), centerline (CE), edge detection (ED), which are conducted wide range real applications. Also, tasks have symbiotic relationship, i.e., SS determines location edges, CE ED can allow...
Earth vision research typically focuses on extracting geospatial object locations and categories but neglects the exploration of relations between objects comprehensive reasoning. Based city planning needs, we develop a multi-modal multi-task VQA dataset (EarthVQA) to advance relational reasoning-based judging, counting, analysis. The EarthVQA contains 6000 images, corresponding semantic masks, 208,593 QA pairs with urban rural governance requirements embedded. As are basis for complex...
Although tremendous strides have been made in facial expression recognition(FER), recognizing expressions non-frontal views remains an open challenge due to the limited access large scale training data with various poses. To make full use of data, we propose a novel multi-channel pose-aware convolution neural network (MPCNN) that consists three parts: feature extraction, jointly multi-scale fusion, and recognition. The extraction part has 3 sub-CNNs it learns convolutional features from...