A5100458768- Remote-Sensing Image Classification
- Cryospheric studies and observations
- Soil Moisture and Remote Sensing
- Remote Sensing and Land Use
- Climate change and permafrost
- Remote Sensing in Agriculture
- Land Use and Ecosystem Services
- Precipitation Measurement and Analysis
- Microbial Applications in Construction Materials
- Remote Sensing and LiDAR Applications
- Climate variability and models
- Urban Heat Island Mitigation
- Synthetic Aperture Radar (SAR) Applications and Techniques
- Anomaly Detection Techniques and Applications
- Image Retrieval and Classification Techniques
- Microwave Imaging and Scattering Analysis
- Advanced Image and Video Retrieval Techniques
- Machine Learning and Data Classification
- Image Processing Techniques and Applications
- Calcium Carbonate Crystallization and Inhibition
- Grouting, Rheology, and Soil Mechanics
- Plant Water Relations and Carbon Dynamics
- Domain Adaptation and Few-Shot Learning
- Geophysics and Gravity Measurements
- Adversarial Robustness in Machine Learning
Nanjing University
2014-2025
Xidian University
2023-2024
China Telecom
2023-2024
China Telecom (China)
2023-2024
Ministry of Natural Resources
2021-2024
State Forestry and Grassland Administration
2023
Shanghai Jiao Tong University
2023
Central South University of Forestry and Technology
2023
Central South University
2023
University of Science and Technology Beijing
2022
Deep neural networks are widely used for classification. These deep models often suffer from a lack of interpretability---they particularly difficult to understand because their non-linear nature. As result, treated as "black box" models, and in the past, have been trained purely optimize accuracy predictions. In this work, we create novel network architecture learning that naturally explains its own reasoning each prediction. This contains an autoencoder special prototype layer, where unit...
Deep neural networks are widely used for classification. These deep models often suffer from a lack of interpretability -- they particularly difficult to understand because their non-linear nature. As result, treated as "black box" models, and in the past, have been trained purely optimize accuracy predictions. In this work, we create novel network architecture learning that naturally explains its own reasoning each prediction. This contains an autoencoder special prototype layer, where unit...
Change detection (CD) of remote sensing (RS) images has enjoyed remarkable success by virtue convolutional neural networks (CNNs) with promising discriminative capabilities. However, CNNs lack the capability modeling long-range dependencies in bitemporal image pairs, resulting inferior identifiability against same semantic targets yet varying features. The recently thriving Transformer, on contrary, is warranted, for practice, global receptive fields. To jointly harvest local-global features...
In recent years, deep learning models, which possess powerful feature extraction abilities, have achieved remarkable success in the classification of hyperspectral images (HSIs). Nevertheless, a common challenge faced by most including few-shot is scarcity valid labeled samples. To address this issue, we propose cross-domain self-taught network (CDSTN) for image classification. The proposed CDSTN merges domain adaptation and semi-supervised strategy to implement learning, utilizes adequate...
The emerging research line of cross-modal learning focuses on the issue transferring feature representation manner learned from limited multimodal data with labelings to testing phase partial modalities. This is essentially common and practical in remote sensing community when only modal-incomplete are users’ hands due inevitable imaging or access restrictions under large-scale observation scenarios. However, most existing methods have been designed exclusive reliance labeling, which can be...
We investigate the non-identifiability issues associated with bidirectional adversarial training for joint distribution matching. Within a framework of conditional entropy, we propose both and non-adversarial approaches to learn desirable matched distributions unsupervised supervised tasks. unify broad family models as matching problems. Our approach stabilizes learning methods. Further, introduce an extension semi-supervised Theoretical results are validated in synthetic data real-world...
Multilabel remote sensing (RS) image annotation is a challenging and time-consuming task that requires considerable amount of expert knowledge. Most existing RS methods are based on handcrafted features require multistage processes not sufficiently efficient effective. An can be assigned with single label at the scene level to depict overall understanding multiple labels object represent major components. The used as supervised information for annotation, whereas additional exploit...
Abstract Extensive and complex changes in spring vegetation phenology have occurred the Pan‐Arctic over last several decades. However, role of snow cover at start growing season (SOS) under different climatic conditions remains unclear. Therefore, we compare effects four indicators on SOS from 1982 to 2015 based long‐term remote sensing data found that end date (SCED) is main indicator affecting SOS, with advancing 0.56 days for each 1‐day advance SCED, explaining 12%–90% variability 63%...
Abstract Ongoing changes in snow cover significantly affect vegetation productivity, but the actual effect of remains unclear due to a poor understanding its lagged effect. Here, we used multisource datasets investigate on productivity Northern Hemisphere ( > 40°N) ecosystems from 2000 2018. We found widespread 40%, P < 0.05) growing season (mean ~73-day lag). The was underestimated by over 10% areas without considering regional time differences. A longer generally occurred warm and...
Recently, the field of hyperspectral image (HSI) classification has witnessed advancements with emergence deep learning models. Promising approaches, such as self-supervised strategies and domain adaptation, have effectively tackled overfitting challenges posed by limited labeled samples in HSI classification. To extract comprehensive semantic information from different types auxiliary tasks, which view problem multiple perspectives, efficiently integrate tasks into a single network, this...
This study compared three broadband emissivity (BBE) datasets from satellite observations. The first is a new global land surface BBE dataset known as the Global Land Surface Satellite (GLASS) BBE. other two are North American ASTER Emissivity Database (NAALSED) and University of Wisconsin Infrared (UWIREMIS) BBE, which were derived independent narrowband products. Firstly, NAALSED was taken reference to evaluate GLASS UWIREMIS more close with bias root mean square error (RMSE) −0.001 0.007...
Sea surface salinity (SSS) plays an important role in global water cycle. In recent years, satellite based remote sensing has proven to be a promising approach for SSS observation. A new payload concept, named MICAP (microwave imager combined active and passive), been introduced this paper. is suit of active/passive instrument package, which includes L/C/K band one-dimensional MIR interferometric radiometer) L-band DBF (digital beamforming) scatterometer, sharing parabolic cylinder...
This paper presents a dynamic classifier selection approach for hyperspectral image classification, in which both spatial and spectral information are used to determine pixel's label once the remaining classified pixels' neighborhood meets threshold. For volumetric texture feature extraction, gray level co-occurrence matrix is used; minimum estimated abundance covariance-based band used. Two remote sensing datasets, HYDICE Washington DC Mall AVIRIS Indian Pines, employed evaluate performance...