The Plant Phenome Journal

Advances in automation, imaging, and artificial intelligence have enabled researchers to capture large volumes of high-quality plant data for understanding crop growth, stress, and genotype-by-environment interactions. While genomics has achieved remarkable throughput, phenotypic data acquisition remains a critical bottleneck for accelerating crop improvement and biological discovery. To address this challenge, an integrated multispectral phenotyping framework was developed using imagery from the Texas A&M AgriLife Precision Automated Phenotyping Greenhouse, a fully controlled facility designed for reproducible plant monitoring throughout the entire growth cycle of most crops. The framework expands the Plant Growth and Phenotyping (PGP v2) dataset and establishes a standardized system for continuous image acquisition, segmentation, deep feature extraction, and temporal analysis across multiple crop species. The project was organized around five coordinated areas: Administration and Coordination, Imaging and Sensor Operations, Data Processing and Management, Artificial Intelligence and Analytics, and Plant Science and Discovery. This structure ensured consistent data quality, version-controlled workflows, and communication across disciplines. The analytical pipeline integrates pseudo-RGB generation, deep learning-based detection and segmentation, image stitching, and temporal (longitudinal) tracking to isolate individual plants and analyze changes in morphology, spectral reflectance, and texture over time. Beyond technical innovation, the framework provides a replicable model for interdisciplinary collaboration and administrative integration in plant phenomics. The combined dataset, workflow, and management framework enable scalable, reproducible, and data-driven plant science research that bridges engineering and biological discovery. Plain Language SummaryTemporal imaging of plants in controlled environments helps scientists better understand growth and biological processes. However, analyzing large volumes of images has been limited by a lack of automated tools. Multispectral imagery captures additional information about plant pigments, structure, and stress beyond standard color images. We developed an automated analysis pipeline that identifies individual plants, tracks their growth over time, and measures traits such as height, area, shape, texture, and vegetation indices. Using artificial intelligence, the system efficiently processes thousands of images to provide consistent and repeatable measurements. By integrating engineering and plant biology, this work supports data-driven decisions for crop improvement and agricultural research.

There is significant enthusiasm about the potential for hyperspectral imaging to document variation among plant species, genotypes or growing conditions. However, in many cases the application of hyperspectral imaging is performed in highly controlled situations that focus on a flat portion of a leaf or side-views of plants that would be difficult to obtain in field settings. We were interested in assessing the potential for applying hyperspectral imaging to document variation in genotypes or abiotic stresses in a fashion that could be implemented in field settings. Specifically, we focused on collecting top-down hyperspectral images of maize seedlings similar to a view that would be collected in a typical maize field. A top-down image of a maize seedling includes a view into the funnel-like whorl at the center of the plant with several leaves radiating outwards. There is substantial variability in the reflectance profile of different portions of this plant. To deal with the variability in reflectance profiles that arises from this morphology we implemented a method that divides the longest leaf into 10 segments from the center to the leaf tip. We show that using these segments provides improved ability to discriminate different genotypes or abiotic stress conditions (heat, cold or salinity stress) for maize seedlings. We also found substantial differences in the ability to successfully classify abiotic stress conditions among different inbred genotypes of maize. This provides an approach that can be implemented to help classify genotype and environmental variation for maize seedlings that could be implemented in field settings. Significance StatementThis study describes the importance of using spatial information for the analysis of hyperspectral images of maize seedling. The segmentation of maize seedling leaves provides improved resolution for using hyperspectral variation to document genotypic and environmental variation in maize.

AO_SCPLOWBSTRACTC_SCPLOWFlowering time is a critical phenological trait in maize (Zea mays L.) breeding programs. Traditional measurements for assessing flowering time involve semi-subjective and labor-intensive manual observation, limiting the scale and efficiency of genetics and breeding improvement. Leveraging unoccupied aerial system (UAS, also known as UAVs or drones) technology coupled with convolutional neural networks (CNNs) presents a promising approach for high-throughput phenotyping of tasseling in maize. Most CNN image analysis is overly complicated for simple tasks relevant to plant scientists. Here a methodology for extracting tasseling from RGB imagery using a CNN-based approach was applied to 220 hybrids and 30 test lines grown in eight diverse environments (Wisconsin and Texas, U.S.A.) then validated through an unrelated set of hybrids. Overall accuracies of .946, .911, .985, and .988 were obtained for classifying maize images with or without tassels from College Station, TX in 2020; College Station, TX in 2021; Arlington, WI in 2021; and Madison, WI in 2021 respectively. By employing deep learning techniques, larger volumes of phenotypic data can be processed enabling high-throughput phenotyping in breeding programs. Although large datasets are required to train CNN models, the proposed methodology prioritizes simplicity in computational architecture while maintaining effectiveness in identifying flowered maize across diverse genotypes and environments.

Continuous, high-frequency monitoring is essential to capture rapid phenological transitions and dynamic crop responses to the environment. However, most phenotyping platforms lack the temporal resolution and automation required for consistent, season-long trait assessment. This study introduces AGIcam, an open-source IoT camera system for automated and continuous in-field plant phenotyping and yield prediction. The platform integrates solar-powered Raspberry Pi units with a modular software stack, comprising Node-RED, InfluxDB, Grafana, and Microsoft Azure, for automated data acquisition, transfer, and visualization. In the 2022 growing season, 18 AGIcam systems were deployed in spring and winter wheat breeding trials, maintaining an uptime of over 85% while capturing frequent RGB and NoIR imagery. Time-series vegetation indices derived from these images were used to predict yield using random forest and Long Short-Term Memory (LSTM) models. The LSTM approach achieved the highest accuracy approximately one week after heading, with mean prediction errors of 3.41% for spring wheat and 1.62% for winter wheat. These results highlight the potential of IoT-based platforms such as AGIcam to enable real-time, scalable, and effective phenotyping solutions for data-driven crop improvement.

Tuber size, shape, colorimetric characteristics, and defect susceptibility are all factors that influence the acceptance of new potato cultivars. Despite the importance of these characteristics, our understanding of their inheritance is substantially limited by our inability to precisely measure these features quantitatively on the scale needed to evaluate breeding populations. To alleviate this bottleneck, we developed a low-cost, semi-automated workflow to capture data and measure each of these characteristics using machine vision. This workflow was applied to assess the phenotypic variation present within 189 F1 progeny of the A08241 breeding population. Our results provide an example of quantitative measurements acquired using machine vision methods that are reliable, heritable, and can be used to understand and select upon multiple traits simultaneously in structured potato breeding populations.

Fusarium head blight (FHB) in wheat is an economically important disease, which can cause yield losses exceeding 50% and the causal pathogen that infects spikes produces harmful mycotoxins. Breeding for host resistance remains the most effective disease control method; but time, labor, and human subjectivity during disease scoring limits selection advancements. In this study we describe an innovative, high-throughput phenotyping rover for capturing in-field RGB images and a deep neural network pipeline for wheat spike detection and FHB disease quantification. The image analysis pipeline successfully detects wheat spikes from images under variable field conditions, segments spikes and diseased tissue in the spikes, and quantifies disease severity as the region of intersection between spike and disease masks. Model inferences on an individual spike and plot basis were compared to human visual disease scoring in the field and on imagery for model evaluation. The precision and throughput of the model surpassed traditional field rating methods. The accuracy of FHB severity assessments of the model was equivalent to human disease annotations of images, however individual spike disease assessment was influenced by field location. The model was able to quantify FHB in images taken with different camera orientations in an unseen year, which demonstrates strong generalizability. This innovative pipeline represents a breakthrough in FHB phenotyping, offering precise and efficient assessment of FHB on both individual spikes and plot aggregates. The model is robust to different conditions and the potential to standardize disease evaluation methods across the community make it a valuable tool for studying and managing this economically significant fungal disease.

Maintenance of sufficient healthy green leaf area after anthesis is key to ensuring an adequate assimilate supply for grain filling. Tightly regulated age-related physiological senescence and various biotic and abiotic stressors drive overall greenness decay dynamics under field conditions. Besides direct effects on green leaf area in terms of leaf damage, stressors often anticipate or accelerate physiological senescence, which may multiply their negative impact on grain filling. Here, we present an image processing methodology that enables the monitoring of chlorosis and necrosis separately for ears and shoots (stems + leaves) based on deep learning models for semantic segmentation and color properties of vegetation. A vegetation segmentation model was trained using semi-synthetic training data generated using image composition and generative adversarial neural networks, which greatly reduced the risk of annotation uncertainties and annotation effort. Application of the models to image time-series revealed temporal patterns of greenness decay as well as the relative contributions of chlorosis and necrosis. Image-based estimation of greenness decay dynamics was highly correlated with scoring-based estimations (r {approx} 0.9). Contrasting patterns were observed for plots with different levels of foliar diseases, particularly septoria tritici blotch. Our results suggest that tracking the chlorotic and necrotic fractions separately may enable (i) a separate quantification of the contribution of biotic stress and physiological senescence on overall green leaf area dynamics and (ii) investigation of the elusive interaction between biotic stress and physiological senescence. The potentially high-throughput nature of our methodology paves the way to conducting genetic studies of disease resistance and tolerance.

Genomic and phenomic analyses suggest additional heritable phenomic features can improve modeling of important end traits like senescence or yield. Field phenotyping generally uses trait values averaged across individual experimental units (plants or numerous plants within plots), ignoring the full distributional pattern of collected measures. Images of plants or plots, as captured by drones (unoccupied aerial vehicles / UAVs / drones), can be viewed as individual distribution functions that capture biological information. This study introduces and validates distributional data analysis in two crops and experiment types - cotton (Gossypium hirsutum L.) single plant vegetation index (VI) analysis and maize (Zea mays L.) plot-level yield predictions. In both crops, the concept of within-day variance decomposition was demonstrated. In cotton, genotypes exerted significant influences on temporal quantile functions of VIs. Maize yield prediction using distributional data with elastic-net regression indicated improvements in yield prediction between 12.7%-21.6% with quantiles outside the conventionally used median responsible for added predictive power. A novel data visualization method for per-pixel heritability allowed distributional features to be explainable and interpretable. These results have implications for future plant phenomic studies, indicating that distributional data analysis applied across temporal imagery captures novel, heritable, and interpretable biological signal that is lost when working with conventional measures of central tendency such as mean or median summary values of experimental units. SignificanceRepeated aerial imaging of agricultural experiments produces image data sets that capture plant development in high spatial and temporal resolutions. Frequently, images are summarized by measures of central tendency, such as mean or median values. Here, functional data distributional methods were applied to cotton (Gossypium hirsutum L.) and maize (Zea mays L.) image data, capturing more information than standard approaches. Cotton genotypes significantly impacted distributional spectral data while in maize, distributional data enabled more accurate predictions of grain yield versus models trained with median data alone. Distributional data were more explainable by genetics, with novel data visualization techniques able to shine light on specific parts of plant imagery with high and low genetic variance.

Hyperspectral reflectance provides rapid and precise phenotyping of plants in a non-destructive manner both in field and well-controlled settings. The resulting data have been used to devise machine learning (ML) models for paired measurements of different traits in diverse plants and crops. Yet, despite advances in using of hyperspectral data to reliably predict crop traits of interest, there are pressing issues concerning the training of ML models, the aggregation of data from crop field trials, and the generalizability of the models in different prediction settings. We collected hyperspectral reflectance data along with 25 anatomical, gas exchange, and chlorophyll fluorescence traits from 320 recombinant inbred lines of a maize Multi-Parent Advanced Generation Inter-Cross population grown across three consecutive seasons. We use these data to systematically: (1) compare the performance of representative ML models for different traits, including slow fluorescence kinetics whose predictability by hyperspectral data has not yet been investigated, (2) evaluate the ML model performance in prediction scenarios concerning unseen genotypes, unseen seasons, and the combination thereof, (3) investigate the effects of data aggregation of ML model performance. These problems are addressed in a rigorous nested cross-validation setting that provides a template for adequate assessment of performance of ML models for diverse crop traits considering the particularities of the experimental design. Significance statementWe present a comprehensive evaluation of hyperspectral reflectance for predicting 25 physiological traits, including fluorescence kinetics, in maize across three seasons using rigorous nested cross-validation. By comparing ML models, prediction scenarios, and data aggregation strategies, our study reveals trait-specific limits of generalizability and offers a robust framework for deploying hyperspectral data in breeding applications.

Precision farming has been developing with the intention of identifying within field variability to adjust management strategies and maximize end of season yield and profitability and minimize negative environmental impacts. The development of quick, easy, and low cost methods to quantify field level variation is essential to successful implementation of precision agriculture at scale. Temporal plant height and growth rates collected with unoccupied aerial vehicles mounted with red, green, blue sensors have the potential to predict end of season grain yield, which could facilitate mid-season management decisions. Image-based plant height data was collected weekly from commercial maize fields in three growing seasons to assess variation within fields and the relationship with grain yield variation. Plant height, growth rate, and grain yield had variable relationships depending on the time point and growth environment. Models developed using temporal traits predicted grain yield variation within a commercial field up to r = 0.7, though insufficient water affected the prediction accuracy in one field due to the limited representation of drought environments in the model development. In the future, with more data from stress environments, such as drought, this method has potential for high accuracy grain yield prediction across a range of environmental conditions. This study demonstrates the potential of using unoccupied aerial vehicles to derive vegetative growth patterns and model within field variations, and has application in making mid-season management decisions.

Excess moisture (flooding, water logging, etc.) is a major source of crop damage causing catastrophic monetary losses to farmers around the world. Losses from excess water are often more common and costly than those from too little water (i.e., drought). Extreme weather patterns are predicted to increase, increasing the expected frequency of excess moisture events to farmers across the Midwest. Despite its importance, studying the impacts of flooding in the field is challenging due to unpredictability of flooding and fields being rendered inaccessible during flooding. Here, we took advantage of a natural flood experiment to examine the responses, damage, and recovery of diverse maize hybrids. Using drones, we monitored the hybrids before, during, and after flooding and examined the spatial and genetic components associated with post-flood survival.

While whole genome sequencing captures millions of single nucleotide polymorphisms (SNPs) and hyperspectral imaging (HSI) enables non destructive plant phenotyping, integrating these modalities to link genotype to phenotype remains challenging due to their high dimensionality and non linearity. This study presents DeepPheno a deep learning framework that predicts SNP genotypes from HSI data, using model predictability as a proxy for genotype phenotype association. HSI data were acquired from 194 lettuce genotypes under field conditions. HSI data patches (20 x 20 pixels x 224 spectral bands) were used to train a hybrid CNN to predict the variant of a specific SNP. The framework was validated on SNPs with known phenotypic effects (anthocyanin, leaf serration, pale pigmentation), achieving high predictive performance (AUC ranging from 0.806 to 0.935), whereas models trained on randomly shuffled labels performed at chance (mean AUC {approx} 0.51). Extending the workflow to 50 randomly selected putatively neutral SNPs, most yielded low predictability, but two showed high performance (AUC > 0.76), suggesting uncharacterized genotype phenotype links. Explainable AI, including SHAP and Grad CAM, identified relevant spectral and spatial features driving these predictions, particularly the green and red edge wavelengths associated with pigment dynamics and leaf structure. These results establish a framework for understanding complex genotype phenotype interactions in plants and extracting these links from HSI data without predefining the exact trait values. It provides an avenue for high throughput trait discovery and description and extends the integration of image based phenomics with plant genetics.

High-throughput 3D multispectral plant phenotyping platforms generate large volumes of point cloud files, but trait extraction is typically performed by sensor-bundled software whose internal algorithms are not publicly documented, which limits reproducibility and integration into custom research pipelines. Here we present PhytoScan3D, an open-source Python pipeline that extracts morphological and spectral phenotypic traits, spanning plant height, 3D leaf area, digital biomass, convex hull volume, leaf inclination, canopy geometry, NDVI, hue, and vegetation indices, from both PLY and PCD point cloud files generated by Phenospex PlantEye F500 and F600 sensors, and is portable to point clouds from any acquisition platform. PhytoScan3D was validated against HortControl (PhenoSpex) ground-truth measurements on 936 barley (Hordeum vulgare) pot-date observations from the growth chamber trial (20 Norwegian cultivars, 12 scan dates, Septemenr 2025 to January 2026), achieving Pearson r = 0.913 to 0.999 and ratio approximately 1.000 for Plant Height Max, 3D Leaf Area, and NDVI Average. A vectorised mesh face filtering implementation achieved a 120x speed improvement, increasing valid 3D Leaf Area coverage from 0.6% to 100% of files. Cross-format validation on 223 PlantEye F600 PCD files from the ICRISAT LeasyScan platform (four legume species: mungbean, cowpea, lima bean, and common bean; 1,523 plant observations) yielded r = 0.884 against independent cuboid annotation heights. The systematic positive bias (mean +27.2 mm, ratio = 1.44) is attributable to PhytoScan3D computing height from raw point cloud Z-range while cuboid annotations are fitted to segmented plant points only, with the offset consistent across all four species (per-species r = 0.880 to 0.888). Cross-dataset processing of 1,180 PLY files from the Crops3D benchmark (8 species, 3 acquisition methods) confirmed zero extraction errors. PhytoScan3D is available at "github.com/kovimallik/phytoscan3d" under the MIT licence and processes 1,651 files across three independent datasets in under 12 minutes on GPU hardware. HighlightsO_LIPhytoScan3D is the first open-source Python pipeline for batch extraction of phenotypic traits, including plant height, 3D leaf area, digital biomass, convex hull volume, leaf inclination, NDVI, and excess green index, from both PLY and PCD point cloud files generated by Phenospex PlantEye sensors. C_LIO_LIPrimary validation against HortControl ground-truth measurements on 936 barley pot-date observations achieved Pearson r = 0.913-0.999 for Plant Height Max, 3D Leaf Area, and NDVI Average. C_LIO_LIA 120x computational speedup in mesh face filtering (vectorised NumPy vs. set-based loop) increased the coverage of valid 3D Leaf Area extraction from 0.6% to 100% of files. C_LIO_LICross-format validation on 223 PlantEye F600 PCD files from ICRISAT LeasyScan (four legume species, 1,523 plants) achieved r = 0.884 against independent cuboid annotation heights. The systematic +27.2 mm bias reflects a methodological difference (raw Z-range vs. soil-segmented annotations), is consistent and predictable across all four species (per-species r = 0.880-0.888), and is correctable by a single linear factor. C_LIO_LICross-dataset processing of 1,180 PLY files from the Crops3D benchmark (8 species, 3 acquisition methods) confirmed zero extraction errors. C_LIO_LISignificant scan-unit variation was detected for Plant Height Max (F = 5.71, p < 0.001, 2 = 0.138) and Canopy Width X (F = 6.32, p < 0.001, 2 = 0.150), demonstrating the biological utility of extracted traits. C_LI

The flowering date of sunflowers is a crucial trait that significantly influences crop management practices and product placement. Traditional ground methods for data collection are labor-intensive and subjective, requiring field scientists to manually estimate and record data in the field. This trait can be measured by counting the number of days from planting until 50% of plants in each research plot have reached flowering at R5 developmental growth stage. However, this method is time-consuming and may overlook valuable information related to flowering rates and duration. Flowering time of sunflower also can be approximated by counting the number of heads (flowers) across multiple dates. We propose a method for rapidly counting sunflower heads to model flower counts over time and estimate flowering time using RGB images acquired by Unmanned Aerial Vehicles (UAVs). The method developed employs a deep learning model trained to detect sunflower heads from UAV imagery and modeling these counts over time using a logistic function to estimate the 50% flowering date. The experimental results obtained from this method enabled estimation of the flowering date with a high correlation to ground measurements (r > 0.91). Significantly, this approach not only reduces labor but also improves the precision of data collection. Moreover, an increase of 6% in heritability across trials, compared to traditional methods, suggests that our approach contributes to a deeper genetic understanding of flowering dynamics. This includes enhanced insights into the timing and rates of flowering, essential for optimizing breeding strategies and understanding genetic responses to environmental conditions. This innovative approach offers a promising avenue for enhancing the efficiency and accuracy of sunflower phenotyping.

Accurate pre-harvest prediction of crop yield informs variety selection, optimizes management, and accelerates breeding. As potato is the worlds leading non-grain staple, here we evaluate a diverse panel of varieties in a three-year field trial across five European locations. Canopy development and environmental parameters are monitored throughout the growing season using drone-based imaging, in-field sensors and gene expression measurements, while tuber yield and quality traits are quantified at harvest. We show that these data enable the identification of climate-resilient, high-yielding genotypes and support the development of machine learning models that explain over 80% of yield variance in independent test sets. Strikingly, measurements collected within the first two months after planting achieve predictive performance comparable to models trained on full-season data. Model interrogation further shows that simplified five-parameter linear equations capture over 70% of yield variability. Our framework thus demonstrates the potential of integrative field phenotyping and data-driven modeling to improve variety selection across heterogeneous environments. Significance statementThe ability to predict harvest crop yields from pre-harvest measurements can enable farmers and growers to make informed decisions on variety selection and management practices, while breeders can benefit from accelerated breeding cycles. We perform a panel of field trials with potato, the no. 1 global non-grain staple, across varying conditions and locations, recording various growth- and climate-related data, including gene expression, and post-harvest yield and quality of tubers. We demonstrate the potential of the field trial data to facilitate the analysis and selection of best-performing varieties across diverse conditions and locations, and to revolutionize farming by enabling early (already within 2 months) and straightforward (only a couple of key measured variables) yield predictions with high accuracy.

Stand count, the number of plants per unit ground area, and leaf area index (LAI), the ratio of leaf area to ground area, are critical traits for crop research but are traditionally measured using labor-intensive methods. While new sensing technologies are being developed, quantifying improvement in measurement efficiency and data quality, relative to traditional techniques, is lacking. In this study, we use LiDAR to generate 3D scans of corn and soybean plots and evaluate two computational methods: a gap fraction approach to estimate LAI and a persistent homology algorithm to estimate stand count by detecting structural peaks in the canopy. Validation experiments and statistical comparisons of bias and variance demonstrate that LiDAR-derived LAI estimates in corn are comparable in quality to those from established instruments. However, in soybean, the LiDAR method performs poorly, likely due to dense canopies limiting light penetration and structural differentiation. Stand count estimations in corn closely match manual counts, with the added benefit of full-plot coverage and significantly faster data collection. In soybean, stand count estimates are unreliable under dense canopy conditions. These results offer practical guidance for the use of LiDAR in field phenotyping and highlight both its current capabilities and limitations. While a trade-off between speed and precision remains, particularly in high-density canopies, LiDARs scalability and multi-trait potential make it a promising tool for high-throughput breeding programs. Continued improvements in LiDAR hardware and algorithm design may further enhance measurement accuracy and extend applicability across crops and growth stages.

Plain Language SummaryThe lack of low-cost, user-friendly and expedient methods for plant phenotyping challenges researchers ability to efficiently collect accurate phenotypic data in large field experiments. Here, we demonstrate the use of a novel, smartphone-based image capture system and two user-friendly image analysis pipelines (utilizing PlantCV or Biodock AI) to increase the throughput of plant phenotyping in two large, spaced plant populations. We showed that the image capture system collected images of adequate quality for downstream analysis using either the PlantCV or Biodock AI pipeline. Both image analysis pipelines produced phenotype values in line with those obtained using manual image annotation. Together, these results provide researchers with a low-cost, user-friendly image-based phenotyping method that can be widely applied to increase phenotyping throughout in field experiments.

High-throughput phenotyping systems are becoming increasingly powerful, dramatically changing our ability to document, measure, and detect phenomena. Unfortunately, taking advantage of these trends can be difficult for scientists with few resources, particularly when studying nonstandard biological systems. Here, we describe a powerful, cost-effective combination of a custom-built imaging platform and open-source image processing pipeline. Our maize ear scanner was built with off-the-shelf parts for <$80. When combined with a cellphone or digital camera, videos of rotating maize ears were captured and digitally flattened into projections covering the entire surface of the ear. Segregating GFP and anthocyanin seed markers were clearly distinguishable in ear projections, allowing manual annotation using ImageJ. Using this method, statistically powerful transmission data can be collected for hundreds of maize ears, accelerating the phenotyping process.

BackgroundThe prediction of desirable traits in wheat from imaging data is an area of growing interest thanks to the increasing accessibility of remote sensing technology. However, as the amount of data generated continues to grow, it is important that the most appropriate models are used to make sense of this information. Here, the performance of neural network models in predicting grain asparagine content is assessed against the performance of other models. ResultsNeural networks had greater accuracies than partial least squares regression models and gaussian naive Bayes models for prediction of grain asparagine content, yield, genotype, and fertiliser treatment. Genotype was also more accurately predicted from seed data than from canopy data. ConclusionUsing wheat canopy spectral data and combinations of wheat seed morphology and spectral data, neural networks can provide improved accuracies over other models for the prediction of agronomically important traits.

Automatically scoring plant traits using a combination of imaging and deep learning holds promise to accelerate data collection, scientific inquiry, and breeding progress. However, applications of this approach are currently held back by the availability of large and suitably annotated training datasets. Early training datasets targeted arabidopsis or tobacco. The morphology of these plants quite different from that of grass species like maize. Two sets of maize training data, one real-world and one synthetic were generated and annotated for late vegetative stage maize plants using leaf count as a model trait. Convolutional neural networks (CNNs) trained on entirely synthetic data provided predictive power for scoring leaf number in real-world images. This power was less than CNNs trained with equal numbers of real-world images, however, in some cases CNNs trained with larger numbers of synthetic images outperformed CNNs trained with smaller numbers of real-world images. When real-world training images were scarce, augmenting real-world training data with synthetic data provided improved prediction accuracy. Quantifying leaf number over time can provide insight into plant growth rates and stress responses, and can help to parameterize crop growth models. The approaches and annotated training data described here may help future efforts to develop accurate leaf counting algorithms for maize.