SimpleMoE-Tea: A Multimodal Mixture-of-Experts Framework for Predicting Fresh Leaf Quality in Large-Leaf Tea

Fresh-leaf biochemical composition reflects canopy condition, leaf phenotype, plant physiology, soil and weather, but rapid assessment studies commonly use a single sensing source or sample-level validation. We developed SimpleMoE-Tea, a modality-specific mixture-of-experts framework for predicting 30 biochemical indicators of large-leaf tea. A multimodal dataset from 11 tea gardens aligned UAV multispectral canopy features, dual-view RGB leaf images, soil properties, annual agrometeorological variables, chlorophyll and nitrogen measurements, elevation and laboratory assays. Separate experts encoded multispectral, image and structured variables; a sample-dependent gating network weighted their representations before regression. Each biochemical indicator was modelled independently. Nested tea-garden-wise cross-validation estimated hyperparameters and all data-dependent preprocessing from the training gardens and reserved held-out gardens for final testing. Across the 90 indicator–outer-fold evaluations, the descriptive mean R2 was 0.945±0.105, the standardized RMSE was 0.250±0.535 and the standardized MAE was 0.125±0.172. The three-modality model produced the highest mean R2 in the ablation study and performed comparably to the strongest interaction-aware baselines. Ten repeated garden-wise evaluations yielded descriptive macro-average values of R2=0.918, standardized RMSE =0.280, and standardized MAE =0.144. These findings support the integration of complementary canopy, leaf and environmental measurements for biochemical quality estimation in fresh tea leaves, while providing a valuable reference for agricultural multimodal data fusion and quality prediction in other crops.

Authors

Institutions

Publication Details

Journal
Foods
Published
2026-11-19
DOI
https://doi.org/10.3390/foods15224160
Primary Topic
Smart Agriculture and AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

SimpleMoE-Tea: A Multimodal Mixture-of-Experts Framework for Predicting Fresh Leaf Quality in Large-Leaf Tea

Zigao Liu, Wei Chen, Wendou Wu, Bing Zhou et al.
Foods
Smart Agriculture and AI
article

SimpleMoE-Tea: A Multimodal Mixture-of-Experts Framework for Predicting Fresh Leaf Quality in Large-Leaf Tea

Zigao Liu, Wei Chen, Wendou Wu, Bing Zhou, Guoquan Pei, Ziran Xu, Zhengzhao Pi, Defu Liu, Benying Liu, Bi Zhao
article en

Abstract

Fresh-leaf biochemical composition reflects canopy condition, leaf phenotype, plant physiology, soil and weather, but rapid assessment studies commonly use a single sensing source or sample-level validation. We developed SimpleMoE-Tea, a modality-specific mixture-of-experts framework for predicting 30 biochemical indicators of large-leaf tea. A multimodal dataset from 11 tea gardens aligned UAV multispectral canopy features, dual-view RGB leaf images, soil properties, annual agrometeorological variables, chlorophyll and nitrogen measurements, elevation and laboratory assays. Separate experts encoded multispectral, image and structured variables; a sample-dependent gating network weighted their representations before regression. Each biochemical indicator was modelled independently. Nested tea-garden-wise cross-validation estimated hyperparameters and all data-dependent preprocessing from the training gardens and reserved held-out gardens for final testing. Across the 90 indicator–outer-fold evaluations, the descriptive mean R2 was 0.945±0.105, the standardized RMSE was 0.250±0.535 and the standardized MAE was 0.125±0.172. The three-modality model produced the highest mean R2 in the ablation study and performed comparably to the strongest interaction-aware baselines. Ten repeated garden-wise evaluations yielded descriptive macro-average values of R2=0.918, standardized RMSE =0.280, and standardized MAE =0.144. These findings support the integration of complementary canopy, leaf and environmental measurements for biochemical quality estimation in fresh tea leaves, while providing a valuable reference for agricultural multimodal data fusion and quality prediction in other crops.

FoodsVol. 15(22)
Yunnan Agricultural University (CN), Zhaotong University (CN), Yunnan Academy of Agricultural Sciences (CN)
Smart Agriculture and AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon