Tested across eight lettuce types, the system successfully visualizes pigment spatial distribution from individual leaves to full canopies, offering a powerful new tool for precision crop management and physiological monitoring.
Plant pigments are vital indicators of photosynthesis, nutrition, and overall plant health. As a widely consumed leafy vegetable, lettuce contains high levels of photosynthetic pigments essential for growth, flavor, antioxidant capacity, and nutritional quality. Chlorophyll drives light absorption and carbohydrate synthesis, while carotenoids protect tissues from oxidative damage and act as precursors of vitamin A in humans. Conventional pigment quantification relies on solvent extraction or chromatography—accurate, yet destructive and unsuitable for continuous field evaluation. In recent years, hyperspectral sensing and machine learning have enabled non-destructive nutrient detection, but most approaches require complex feature preprocessing, lack cross-scale assessment, and struggle to generalize across varieties. These challenges highlight the need for a rapid, automated, and scalable method for pigment assessment in living plants.
A study (DOI: 10.1016/j.plaphe.2025.100104) published in Plant Phenomics on 12 September 2025 by Liping Chen’s team, Beijing Academy of Agriculture and Forestry Sciences, establishes a high-accuracy, non-destructive method for estimating and visualizing lettuce pigment content using hyperspectral imaging combined with deep learning, enabling efficient physiological assessment and supporting precision agriculture.
In this study, the authors first performed quantitative analysis of pigment content in leaves from eight lettuce types, combined with hyperspectral reflectance measurements and a suite of modeling approaches, to explore how pigments vary within and among varieties and how well they can be predicted from spectral data. They characterized the distributions of Chl a, Chl b, Car, and TPC, analyzed vis–NIR reflectance curves (366–976 nm) and red-edge features, and preprocessed hyperspectral data using MA, SNV and FD1 to enhance spectral peaks and troughs. Machine learning models were systematically built under different combinations of preprocessing, feature selection (including CARS and SPA), dataset partitioning (Random, KS, SPXY), and regression algorithms (PLSR, RF, SVR, ELM); then, an end-to-end deep learning model, LPCNet (CNN + BiLSTM + MHSA), was trained and evaluated. Finally, leaf-level inversion models were extended to canopy reflectance to visualize pigment maps, and canopy pigment statistics were used to compare distribution patterns among lettuce types. The results showed distinct statistical distributions for each pigment: bimodal for Chl a (0.68–0.99 mg/g), unimodal for Chl b (0.29–0.45 mg/g), right-skewed for Car (0.14–0.23 mg/g), and approximately normal for TPC (1.19–1.55 mg/g), reflecting diverse metabolic strategies. Spectral analyses confirmed characteristic absorption troughs at 430–470 nm and 670–690 nm and a pronounced red edge, with preprocessing improving signal stability. Optimal machine learning combinations achieved high R² values (up to ~0.91), but LPCNet further improved accuracy and robustness, reaching R² up to 0.94 on prediction sets and lower MAE. False-color canopy maps revealed spatial gradients of Chl a, Chl b, Car, and TPC and consistent patterns within, but clear differences among, lettuce types, linking pigment distribution to genetic background, leaf morphology, and light adaptation strategies, and providing a solid basis for breeding and cultivation optimization.
The study establishes a fast, non-invasive pigment estimation pipeline suitable for high-throughput phenotyping, smart greenhouses, and digital farm monitoring. Real-time pigment maps enable growers to evaluate photosynthetic efficiency, diagnose nutrient stress early, optimize fertilizer strategies, and support breeding selection based on pigment traits. For research, the system offers a new lens for exploring pigment metabolism, light-response mechanisms, and plant biochemical variation under environmental change.
###
References
DOI
10.1016/j.plaphe.2025.100104
Original Source URl
https://doi.org/10.1016/j.plaphe.2025.100104
Funding information
This work was partially supported by the Beijing Rural Revitalization Agricultural Science and Technology Project (NY2401040025), the National Key R&D Program (2022YFD2002300), and the Construction of the Collaborative Innovation Center of Beijing Academy of Agricultural and Forestry Sciences (KJCX20240406).
About Plant Phenomics
Plant Phenomics is dedicated to publishing novel research that will advance all aspects of plant phenotyping from the cell to the plant population levels using innovative combinations of sensor systems and data analytics. Plant Phenomics aims also to connect phenomics to other science domains, such as genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics, and computer sciences. Plant Phenomics should thus contribute to advance plant sciences and agriculture/forestry/horticulture by addressing key scientific challenges in the area of plant phenomics.