Research program
From sensing crops to designing crops.
Four layers turn multi-source observations into a crop that is measurable, explainable, predictable and, eventually, designable. Each layer below links to the projects, papers, notes and tools that support it.
- 2
- projects
- 2
- publications and software
- 11
- research notes
Architecture
Four layers of crop intelligence
My PhD built Layers I–II. My postdoc focuses on the jump from II to III: making crop state not only observable and explainable, but projectable.
Projects
Selected projects
Predicting leaf BRDF from phenotypic traits
Leaf directional reflectance can be predicted from measurable traits, and the optical diversity this reveals changes how light is distributed inside a simulated canopy.
MCTP: a multi-modal crop phenotyping workspace
One desktop workspace gives hyperspectral, LiDAR, RGB and thermal processing a shared entry point and export convention, while keeping each modality's processing transparent and tunable.
Notes
Research notes
- PhenoHUB: A Mobile Toolkit for Digital Plant PhenotypingIDigitize
A January 2026 snapshot of a WeChat Mini Program combining field utilities, weather queries, image tools, and experimental AI assistants.
- Botanical Extract AI Pro: An Experimental Background-Isolation WorkflowIDigitize
An experimental web and batch workflow for placing plants on white backgrounds with multimodal image models, including scientific and privacy limitations.
- MCTP: A Multi-Modal Crop Phenotyping WorkspaceIDigitize
A desktop workspace for hyperspectral, LiDAR, RGB, and thermal crop-phenotyping workflows, with clear boundaries between shared UI and cross-modal fusion.
- Predicting Leaf BRDF from Phenotypic TraitsIIUnderstand
A peer-reviewed framework combining directional spectroscopy, BRDF fitting, phenotypic traits, ensemble learning, and canopy ray tracing in four species.
- Local Image Quantification in Python: An Experimental, Auditable WorkflowIDigitize
A compact OpenCV workflow for segmenting isolated biological samples, exporting pixel and calibrated size descriptors, and documenting the validation required before scientific use.
- Hunyuan3D-1 for Plant Images: A Reproducible Exploration GuideIDigitize
A version-specific, evidence-aware workflow for generating exploratory plant meshes with Hunyuan3D-1 and validating why generative 3D output is not automatically a phenotype measurement.
- UAV 3D Crop Phenotyping: From CCO Acquisition to Validated TraitsIDigitize
A field-to-analysis workflow for Cross-Circular Oblique UAV acquisition, SfM reconstruction, spatial referencing, point-cloud phenotyping, and honest model validation.
- Root Quantify: Interactive Root Image Preprocessing in PythonIDigitize
A practical guide to using Root Quantify for polygon ROI selection, background correction, binary-mask cleanup, and organized export before downstream root analysis.
- Canopy Photosynthesis from 3D Plant Models — A Conceptual WorkflowIIUnderstandIIIPredict
A validation-first research framework connecting multi-view reconstruction, plant geometry, radiative transfer, leaf physiology, and canopy-scale uncertainty.
- DJI P4 Multispectral to Plot-Level Traits with WebODM and QGISIDigitize
A validation-first workflow for processing DJI P4 Multispectral imagery in WebODM, checking band metadata in QGIS, and extracting plot-level vegetation features.
- Turntable Photogrammetry for Potted Cotton in a Growth ChamberIDigitize
An experimental, validation-first protocol for acquiring and reconstructing multi-view images of potted cotton plants in a controlled environment.
