Research Interests

Urban sustainability, Environmental remote sensing, Land use and land cover mapping, Multi-source data fusion, Time series analysis, Machine learning, and Deep learning

Research Experiences

Ongoing research projects

[1] Evaluating the visual accessibility of urban greenspaces from different building heights on a global scale

• Develop an efficient method to calculate the green view index (GVI) of different building heights.
• Evaluate the visual accessibility of urban greenspaces across different cities worldwide.

[2] Generating a worldwide high spatial resolution hourly land surface temperature product

• Develop a high spatial resolution, hourly LST data generation framework using ECOSTRESS LST data, as well as multi-source data.
• Apply the framework globally to generate a valid product.

Several previous projects

[1] Research on Segment Anything Model (SAM)-Assisted remote sensing crop mapping

Enhancing crop mapping through an automated sample generation framework based on SAM

• Evaluate the performance of SAM for crop parcel segmentation using medium-resolution satellite imagery, such as Sentinel-2 and Landsat-8.
• Develop a novel automated sample generation framework based on SAM.
• Assess the effectiveness of the framework in Henan Province of China and southern Ontario of Canada.
• Relevant code can be found at SAM-CropSampleGeneration. Related publication: Sun et al., 2024.

A weakly supervised learning method based on SAM for crop mapping

• Use adapters to finetune SAM for crop parcel segmentation in Sentinel-2 images.
• Generate high-quality pseudo labels through finetuned SAM and weak annotations, replacing the labor-intensive process of obtaining pixel-level annotations.
• Apply pseudo labels to train a fully supervised segmentation model to conduct crop mapping.
• Relevant code can be found at SAMWS. Related publication: Sun et al., 2024.

[2] Large-scale crop mapping with multi-source satellite images using a spatiotemporal datacube-based deep learning framework

• Develop a datacube-based framework to conduct large-scale crop mapping.
• Adopt a novel sample extraction technique based on spatiotemporal datacube.
• Fuse GF-1 and Sentinel-2 multi-temporal images by early and late fusion strategies.
• Related publication: Sun et al., 2024.