Current phenotyping technologies, whether based on cameras, LIDAR or hyperspectral imaging, are capturing mainly 3D surfaces and are limited, to penetrate into growth media and internal tissue and structures. 3D X-ray computed tomography (CT) alleviates many of these shortfalls by enabling a non-destructiv visualization of optically inaccessible plant structures, allowing for the 3D reconstruction and measurement of objects at high resolution and high throughput. We will present a range of fully automated, industrially validated 3D X-ray CT based technologies, that enable to visually and quantitively follow up in 4D the entire plant development cycle from flowers/ears to seeds to germinating seedling in filter paper to plants and root structures in soil. We will emphasize the non-destructive fully-automated 3D phenotyping of seeds and the resulting germinating seedlings including their internal organs in filter paper across time, i.e. in 4D, at a current throughput of 200 seeds/min and 25 seedlings/min, respectively.The presented technologies, being universally applicable across plant and crop species, allow for the quantitative, objective and reproducible assessment of morphological seed and seedling traits in 4D. They provide powerful tools to investigate any influence, whether genetic, environmental or treatment-related on seed quality and the germination capacity, vigor and 3D phenotype of the resulting seedling over large samples as big data. We will present the technologies and data on traits such as seed quality, seedling development, degree of abnormalities, germination capacity and vigor across different crop types.

In the Chicago region, common buckthorn and bush honeysuckle stand out as dominant invasive species, accounting for over 40% of the regional forest coverage. Their proliferation leads to the formation of dense thickets that hinder sunlight penetration, resulting in diminished native plant diversity in the understory. Accurate detection of these species across this region is essential for the effective management of these invasives. Airborne LiDAR datasets show strong potential in detecting the distinctive structural patterns created by these thickets. Previous studies have utilized multi-temporal spectral imagery to track the phenological shifts of invasive species. However, in the Chicago region, it is difficult to employ multi-temporal spectral imagery due to substantial cloud cover during early spring and late autumn. Consequently, our research adopted the use of dense airborne LiDAR datasets specific to the Chicago area. Preliminary results indicate that invaded plots manifest less complex vertical structure, higher vegetation area index in the subcanopy, and lower NDVI values compared to their non-invaded plots. Notably, LiDAR-derived metrics surpass NDVI-based ones in estimation. Using binomial logistic regression (with an AUC of 0.97), we assessed the presence and absence of these invasive species across Chicago's forests, achieving an accuracy rate of 0.92. Alarmingly, our findings suggest that these invasive species have affected over 75% of the forest patches in Chicago. In summary, our research highlights the important role of LiDAR datasets in regional-scale invasive species detection.

Northern corn leaf blight (NCLB), caused by the fungus Setosphaeira turcica, is one of the most devastating corn diseases in the U.S. and has reduced Indiana yields for the last decade. NCLB can cause significant yield loss if conditions are favorable for early disease onset near tassel/silk (VT/R1) growth stages. Therefore, early detection and timely disease management are essential to mitigate yield loss and optimize fungicide applications. Remote sensing technologies have broad applications in plant phenotyping areas, but they present specific challenges in the early detection of diseases, particularly for diseases where lesions first appear in the lower crop canopy. While laboratories can provide more accurate results than remote sensing, their throughput is limited. To achieve higher throughput and more precise measurements for improved early disease detection, a portable digital microscope prototype has been developed. Distinct from other portable microscopes, this device can detect pathogens and directly monitor their growth on live plant tissue due to its 1 nm spatial resolution. Moreover, it is built with a confocal configuration to capture high-resolution reflective images, thereby outperforming traditional microscopes in throughput. This feature also permits repetitive non-destructible measurements on a single leaf and accommodates leaves of varying thicknesses. By analyzing time-series images of infected leaves, this portable digital microscope has proven its ability to detect spores and germ tubes of the pathogen before symptoms are visibly apparent. It has the potential to effectively bridge the gap between precision and throughput, providing critical insights into disease management.

UAS-based image analytics has been deployed to expedite the plant phenotyping and replace laborious manual notetaking but is limited for global validation in all field conditions. Plot-level metrics is essential for plant phenotyping on many plots and extracted by defining a region of interest (ROI) of the field boundary and processing sub-ROIs aligned with rows and columns of the total number of plots, called gridding. Gridding is offered by commercial software but is limited to upright rectangular fields. When UAS tile images are stitched, an orthomosaic image is georeferenced to make the image top to north, whereas the field orientation is often off the north. Due to the misaligned orientation, the gridding process requires a preprocess of image rotation to align the grid onto the field boundary, which creates resampling errors and takes laborious multiple adjustments to precisely align sub-ROIs with plots across the field. To address this issue, an open-source software was developed to generalize the gridding method and provide a quick extraction of plot-level metrics without the image rotation. Adaptive gridding algorithm is to rotate the grid by applying geometry of a rectangle in a circle that keeps right angles. Metrics of the rotated ROI is calculated by geofencing pixels in the ROI for segmentation, filtering, masking, and clustering. The open-source software with adaptive gridding allows the end-users to process their UAS images for high throughput phenotyping in an effective manner without understanding details of image processing.

Early detection of plant stress is important, particularly in the face of unpredictable climates. Conventional methods of phenotyping, while useful, lack the ability to provide non-destructive early plant stress detection, a capability provided by remote sensing technologies. The aim of this study was to determine the effect of planting date on various commercial maize hybrid growth and development, focusing on changes in spectral reflectance and vegetation indices related to plant health. An experimental study was set up in Pretoria, South Africa, encompassed planting dates spanning from November 2022 (Early) over December 2022 (Optimum) and January 2023 (Late) to February (Very late). Weekly UAV (unmanned aerial vehicle) flights starting four weeks after planting were conducted with the DJI M210 carrying the Micasense Altum multispectral sensor. Preliminary findings identified a higher spectral reflectance of maize, in the visual, red-edge, and near-infrared bands at optimum planting date as compared to the earlier and later planting. These shifts in spectral reflectance were also reflected in increased values of the Simple Ratio Index (SI), Green Ratio Vegetation Index (GRVI), and Normalized Pigment Chlorophyll Index (NPCI) in the earlier planting dates, when contrasted with the later planting dates. The findings confirm that planting date impacts maize health, and that this can be detected through remote sensing methods. This research demonstrates that UAV technology is a feasible alternative to fixed plant phenotyping platforms, especially in the African context where infrastructure stability remains a major challenge.

South Africa is dependent on dryland farming, causing crops to be sown only once the rainy season begins. With climate change altering weather patterns, unpredictable rainy seasons may lead to cash crops such as maize and soybean being planted later, which makes them susceptible to extreme temperatures and drought. Farmers often switch to planting sunflower (Helianthus anuus L.) under these conditions because of its shorter rotation time and hardier nature. For this reason, sunflower tends to be planted outside of its optimal planting window. To understand the effects of different planting dates on sunflower growth and development, a planting date trial was established in Potchefstroom under the Phenospex FieldScan system. Low-and high-density monthly plantings were sown from October to March and scanned three times a day (RGB, NIR, Laser). Values are extracted to calculate plant growth and health indices over the season for each planting date. The aim of this project is to develop a Shiny application that can analyse the Phenospex data outside of the platform. Phenospex software is expensive and impractical to use in the South African context with unstable infrastructure because it requires the sever to be on in one location for analysis in a different location. The Shiny app will allow Phenospex data to be analysed offline, offering more flexibility in an African setting.

This work presents a methodology for creating digital twins of root system architecture (RSA) that can be used for studying the phenotypic variation in RSA. Growing populations demand increased global food production. To sustainably support this increase, crops must be developed to flourish in nutrient-depleted soils. Since the effectiveness of nutrient uptake is determined by plant rooting system dynamics, much focus has been placed on studying RSA across species and varieties. A particularly effective tool for studying RSA in 3D has been X-ray computer tomography (CT). However, this technology is cost prohibitive and cannot model field-grown samples. A far more cost-effective technology is close-range photogrammetric scanning, which uses multiple 2D images to reconstruct 3D point clouds of RSA. This project develops a point cloud processing pipeline that takes high-density point clouds of soybean rooting structures and generates 3D RSA models. These digital twins are then used as the basis for analyzing the phenotypic variation of the geometric and biometric features of the RSA. We believe this digital twin construction and analysis pipeline will increase the impact of RSA research in support of sustainable food production in nutrient-depleted areas of the world.

This study explores cross-species variations in structural metrics derived from individual tree lidar scans and analyzes the structural characteristics of trees at both the individual and plot levels. Understanding the species composition of forests is vital for assessing ecosystem health and biodiversity. For this reason, much research has been dedicated to distinguishing between species through remotely sensed datasets; however, the goal of remote species identification remains unrealized. Progress has been hindered by two key challenges: 1) the lack of precisely georeferenced ground validation data, and 2) insufficient consideration of 3D environmental factors such as terrain, understory layers, and the structural entanglement of natural forests. In this study, we address these challenges by compiling a ground validation dataset comprised of precisely high-density laser scans at 60 forest inventory plot locations. Within this dataset, each of the 1200 trees is segmented and paired with a manually verified species label. Using this comprehensive dataset, we first isolate individual trees and investigate cross-species variations in lidar feature metrics. Subsequently, we examine how these structural features vary within the context of the micro-environment of their respective plots. Our analysis also explores how factors such as terrain and neighboring species impact structural feature variance. Our findings suggest that detectable structural features hold significant potential in accurately assessing biodiversity and determining the species composition within forests.

High spatial resolution remote sensing and machine learning have improved the accuracy and affordability of high-throughput phenotyping. This paper presents an innovative approach to yield prediction, leveraging multimodal networks and integrating remote sensing data from two sensing technologies. We explore the synergy among different types of remote sensing data, meteorological data, and management practices employing advanced machine learning techniques to enhance accuracy and reliability in yield predictions.This study focuses on three experiments, one in 2021 and two in different environments with distinct management practices in 2022. Hyperspectral, LiDAR and weather-related features served as initial inputs to the LSTM recurrent neural network-based yield prediction models; management practices related categorical variables were concatenated after the time series output from the attention network. In each experiment, 80% of the data was used for training the model and 20% for testing, investigating three deep learning networks:Traditional vanilla stacked LSTM network.Stacked LSTM Network with a temporal attention mechanism.Multi-modal network for the different remote sensing modalities.The attention weights of each time-step were evaluated to determine the importance of each date;All models produced good predictions, but the attention mechanism coupled with the multi-modality network showed the effectiveness of combining multimodal remote sensing data optimally throughout the season and deep learning algorithms in optimizing agricultural decision-making processes. The Attention networks also provided increased interpretability over the growing season, showing flowering time to be the most critical time for the models, which is consistent with field-based trials.

PlantCV is a Python-based image analysis tool that lowers the barrier to entry for complex image analysis workflows in plant phenotyping. To provide support for subsequent analysis steps of measured trait data we have developed pcvr, an R package to assist in common plant phenotyping analyses. The goal of pcvr is to make common statistical analyses both easier and more consistent and to lower the barrier to entry for useful Bayesian methods. Here we demonstrate three pieces of a possible analysis covering single value trait analysis, longitudinal modeling, and multi-value trait analysis.

Feeding the world’s population of 9 billion by 2040 is one of the major challenges of the agriculture sector. Wheat (Triticum aestivum L.) is the second most important staple crop, with a global production of 773 million tonnes per year, but the expected yields need to increase by 60% to ensure future food security. Achieving this requires the development of new cultivars with heightened expression of yield-associated traits, such as radiation use efficiency (RUE), which is fundamental to enhancing plant performance and yield. Recently, plant architectural phenotypes involving leaf inclination angle have shown promising traits in improving RUE at the canopy level. Specifically, the erectophile leaf arrangement exhibits a higher yield potential, receiving more even light distribution than the planophile arrangement, which is susceptible to light saturation on the top layer. This study used a mobile robotic phenotyping system with 3D-multispectral laser scanners and hyperspectral cameras. In 2022 and 2023, data were collected from 100 spring wheat canopies at the heading/anthesis and booting/anthesis stages, respectively. Using 3D data, we estimated canopy tangency angles, identified two architectural phenotypes, and incorporated them into one-dimensional (1D) convolutional neural networks (CNN) to predict canopy-based RUE. Canopy architectural phenotypes in CNN models improved the prediction accuracy of RUE. These findings underscore the potential of canopy architectural traits derived from 3D images as a critical parameter for enhancing RUE predictions in wheat canopies. It could potentially be used in other cereal crops.

Given the changing environmental conditions, understanding internal mechanisms of crops and their interaction with the environment in abiotic stress conditions is very important. To estimate the yield of e.g. wheat, spike and grain traits are used for the selection of resilient genotypes for further breeding programs. Typically, the extraction of these traits is laborious and destructive. With automated X-ray technology a non-destructive screening of spikes can be conducted faster and more accurate. Furthermore, the technology allows extracting additional traits such as center of mass and aspect ratio of each grain, together with the position of the grain within the spike.We present the analysis of 203 wheat accessions carried out using a portable computed-tomography system (CT). The wheat plants were exposed to either drought or combined drought and heat stress. With the EarS algorithm, spike traits were analyzed with an accuracy of 95-99%. The system scanned 4 spikes simultaneously, which resulted in a scanning-time of 7 minutes per spike. To enable a high-throughput acquisition and comparison of wheat traits, we present the required steps for automation. Thus, together with PhenoKey, we developed an automated CT-system with integrated conveyor belt for The Plant Accelerator®, University of Adelaide. This system examines spikes and grains with high throughput. The system has space for 35 spike holders, each of which can carry 30 spikes. This allows a scanning time of 13 seconds per spike. Combined with an automatic data post processing each grain can be identified and linked to the corresponding genotype.

Understanding and enhancing plant resilience traits in maize is critical for ensuring consistent yields in the face of challenging conditions such as drought, disease, and insect pressure. The adoption of drought-resistant and insect-resistant maize varieties has already proven highly beneficial to growers, sparking further interest in uncovering the precise genetic and phenomic factors associated with increased resilience. This study utilized genetic, phenomic, and environmental data collected by Genomes to Fields teams across diverse regions of the United States. These regions exhibited a wide range of climatic conditions, from optimal to stressful, including drought and heat challenges. Leveraging advanced Machine Learning techniques, the study aimed to predict the impact of drought and heat stress on the anthesis-silking interval, a crucial trait influencing yield, and to identify the most effective combination of phenotypes for enhancing drought resilience in maize.

Corn is a cornerstone of the global food supply, but its vulnerability to unpredictable weather patterns, exacerbated by climate change, poses significant challenges to its consistent production. This study aims to enhance maize breeding efforts by identifying genetic markers associated with plant resilience, including the factors of growth rates, stress resistance, disease resistance, overall productivity, and variation in weather. Leveraging previous years of genetic, phenotypic, and drone imagery data from the maize Genomes to Fields project, we conducted a genetic analysis utilizing location-year datasets with varying disease presence, environmental stressors, and weather conditions. The focus was primarily on the anthesis-silking interval (ASI), an essential metric in drought resistance. Machine learning techniques were employed for efficient large-scale data analysis, hypothesis testing, and result interpretation. The outcomes of this research hold immense value for farmers grappling with seasonal droughts and extreme heat. Cultivating crops with a history of weather resilience can serve as a predictive tool, mitigating supply chain disruptions and food shortages, ultimately stabilizing food prices.

Maximizing crop yield is a central goal in agriculture and plant breeding. Several phenotypic traits, including leaf chlorophyll content and canopy greenness, can help predict yield across different varieties. In this study, we aimed to quantify the relationship between leaf chlorophyll concentration, canopy greenness, and crop yield within the Genomes to Fields trials. Our goal was to assess the correlation and enhance predictive modeling between these factors, evaluating the effectiveness of canopy cover and spectral reflectance in predicting yield. We accomplished this by collecting chlorophyll data from multiple leaves within each field plot and utilizing imagery captured by unoccupied aerial systems. This image data provided RGB and multispectral reflectance information on a per-plot basis throughout the growing season, allowing us to model patterns linking chlorophyll concentration and yield for each plot. We integrated data from these two methods with genotypic data from the same maize varieties to explore the fundamental relationships between chlorophyll levels, canopy greenness, and crop yield.

Plant phenotyping technologies have been developing rapidly over the last 2 decades. Plant sensors are becoming more accurate, faster, and easier to use. However, there are still bottleneck issues in plant sensing, including the changing environmental conditions, the plant’s diurnal activities, and the complicated Genotype-by-Treatment-by-Environment interactions. These issues keep plant phenotyping difficult and limit further application of the sensor technologies in precision agriculture. The Purdue Ag Engineers have been working innovatively to develop the next generation sensor technologies to address these issues. In this presentation, Dr. Jin will firstly introduce the four automatic high-throughput phenotyping facilities recently developed at Purdue and explain why they were built and how they’ve been successfully used in detecting diseases, nutrients deficiencies, and chemical damages. He will then introduce ABE’s most recent plant sensor technologies, including the 2021 Davidson Prize winner, LeafSpec, and the new drone and ground-based robots for automatic field phenotyping in different scenarios. Dr. Jin will also briefly share some discovery stories in plant phenotyping, such as how the crop images diurnal variances, and how to leverage the advantages of remote sensing and proximal sensing. 

Carrot (Daucus carota L.) can take up high concentrations of toxic heavy metals like cadmium (Cd) and store them in their edible taproots, leading to food safety risks.  Cd poisoning in humans is strongly linked to damage to the liver, lungs, and bones, as well as prostate, kidney, pancreatic, and testicular cancer.  Cd may also be detrimental to carrot plant growth and nutrient content. A potential way to address this challenge is to select for carrot varieties with lower uptake, however, this is difficult because plants generally exhibit few visible signs of Cd stress. The goal of this study was to determine if hyperspectral imaging can be used as a tool to screen for carrot breeding lines that can withstand Cd stress and restrict uptake, thereby accelerating breeding efforts. To accomplish this goal, we grew 6 carrot breeding lines previously shown to differ in their tendency to accumulate toxic heavy metals like Cd at Purdue’s Ag Alumni Phenotyping Facility (AAPF).  Carrots were either treated or not with cadmium chloride (CdCl2), and imaged with red-green-blue (RGB) and hyperspectral cameras throughout their growth. After 2 months, plants were destructively harvested and Cd and other elements were quantified using ICP.Results of this study verified differences in the uptake of Cd and other elements between the carrot lines, and several reflectance values and vegetative indices have potential to detect Cd stress in carrot plants.

As one of the largest supplied grain crops, corn plants often require a significant amount of nitrogen fertilizer for optimal yield. However, excessive fertilizer usage can lead to adverse environmental consequences, especially for the nearby hydrological network. To precisely manage nitrogen application, accurate measurement of corn crop nitrogen deficiency is necessary. Hyperspectral imaging (HSI) techniques are widely applied in plant phenotyping to effectively measure plant traits caused by biotic or abiotic stresses. While previous HSI processing methods primarily focus on the overall color change, they rarely analyze the signal from the leaf-level spatial domain. However, early-stage nitrogen deficiency symptoms may not significantly alter the overall color, resulting in limited model performance in such cases. A newly developed HSI device called LeafSpec can scan an entire corn leaf with a high signal-over-noise ratio paired with high spatial-spectral resolution, capturing the detailed color changes at the leaf structure level. This study focused on identifying distinctive nitrogen deficiency indicators using an innovative methodology that applies spectral analysis to the details of leaf venation structures. The study started with developing an automated venation segmentation algorithm to separate a whole corn leaf into structural components. An in-depth examination of the spectral profiles associated with different leaf components introduced a new spatial-spectral index, demonstrating a higher correlation with the nitrogen content data than the averaged spectral indices. The high-resolution spectral-structural features discovered with this method provided new potential to improve the performance of the nitrogen prediction model in terms of both accuracy and robustness.