Understanding root traits is essential to improve water uptake, increase nitrogen capture and raise carbon sequestration from the atmosphere. However, high-throughput phenotyping to quantify root traits for deeper field-grown roots remain a challenge. Recently developed open-source methods use image-based 3D reconstruction algorithms to build 3D models of plant roots from multiple 2D images and can extract root traits and phenotypes. Most of these methods rely on automated image orientation (Structure from Motion)[1] and dense image matching (Multiple View Stereo) algorithms to produce a 3D point cloud or mesh model from 2D images. Until now it is not known how the performance of these methods compares to each other when applied to field-grown roots. We investigate commonly used open-source pipelines on a test panel of twelve contrasting maize genotypes grown in real field conditions in this comparison study [2-6]. We compare 3D point clouds in terms of number of points, computation time, and model surface density. This comparison study will provide insight into the performance of different open-source pipelines for maize root phenotyping, and illuminates trade-offs between 3D model quality and performance cost for future high-throughput 3D root phenotyping.

Phenotypic trait measurements have enabled breeders to link genomic information to phenotypic information and through this enhance crop performance by breeding for superior germplasm. Progress in this area has been hindered by the limited ability to capture agronomic traits of importance at a large field-based scale since traditional methods for measuring phenotypic traits in field are time and labor intensive and are limited in accuracy and consistency when implemented on large scale. Precision phenotyping efforts have enabled researchers at Corteva Agriscience to collect high quality datasets for important traits that programs had been unable to measure accurately or safely in the past. Specific applications of phenotyping technologies and how these have influenced data collection within Corteva breeding programs around the world will be presented. Technology is rapidly evolving, and by using this technology to develop novel precision phenotyping solutions, breeders are able to capture more high-quality data and gain unique insights.

To assist plant scientists, geneticists, and growers to understand crop-environment interactions, plant phenotyping is a powerful tool for improving crop cultivars and developing decision support systems in farm management. Recent trends use LiDAR to capture three-dimensional (3D) information from plants to analyze traits vital to plant growth and development. However, current terrestrial-based 3D analysis methodologies are time and labor intensive and can be a bottleneck when large agricultural fields need to be analyzed. Robotic technologies can be used to accelerate the field-based measurements of relevant plant features and optimize the high-throughput phenotyping process. In this paper, we present a robotic system with a 3D LiDAR and a data processing pipeline for efficient, high-throughput field phenotyping of cotton crops. The robotic system consists of a Husky robotic platform equipped with a FARO Focus 3D laser scanner. The components of the system are integrated under the ROS framework to ensure interoperability and data integrity and availability at any given time. The data processing pipeline involves the data collection, registration, and analysis tasks for measuring crop traits at the plot level—canopy height, volume, and light interception—and estimating yield. This work demonstrates a crop phenotyping platform that leverages two off-the-shelf equipment for the quantitative assessment of cotton plant traits in the field. This methodology can be extended to other agricultural crops contributing to the advancement of plant phenomics.

Crop pest detection and mitigation remains an extremely challenging task for the farmers. Majority of the pest classification and detection techniques rely on supervised deep learning frameworks that require significant human intervention in labeling the input data, thereby making the down-stream tasks tedious. Therefore, this study presents a self-supervised learning (SSL) approach to classifying 12 types of agricultural insect pests from 9549 RGB images, by leveraging the Bootstrap your own latent (BYOL) algorithm. SSL uses minimal labeling and is indifferent to data augmentations or distortions. Hence, latent representations from pretrained SSL networks could be generalized well for downstream tasks like classification or object detection. For desirable classification of the insect images, the greatest challenges observed were: i) large intra-class variation (the same insect was found with different colors and patterns), and ii) complex background with inconspicuous foreground. Hence, to overcome these issues and aid generalizability of the representations learned through BYOL, entropy-guided segmentation (segments based on texture not color), is proposed as input to the SSL network in this study. Both raw and segmented images were separately fed to two independent BYOL SSL networks, i.e., with ResNet18 and ResNet50 architectures as the backbone. The efficacy of the latent representations for downstream applications was assessed using linear evaluation, and subsequently compared with traditional transfer learning outcomes from ResNet18 and ResNet50. The results indicated that, while ResNet50 backbone intuitively performed better in all cases, SSL aided with entropy-based segmentation led to ~94% classification accuracy compared to raw images (with ~90% maximum accuracy).

Canola has a prominent floral signature and requires careful consideration when selecting spectral indices for yield estimation. This study evaluated several spectral indices derived from high-resolution RGB images. A small plot (2.75m x 6m) experiment was conducted at Kernen Research Farm, Saskatoon, where canola was grown under varying row spacings and seeding rates (192 plots). The canopy reflectance was imaged during the flowering period and seed yield was obtained at physiological maturity. Indices were evaluated for accuracy in quantifying canola flowers in high-resolution RGB imagery with within-canopy shadow pixels. Digitalized flower number from the peak flowering date was used to test and validate a non-linear three-parameter asymptotic regression model to simulate canola seed yield. 70 % of the data was used to develop the model, and 30 % was used to validate the model. Model performance was tested with Pseudo-R2, r, MAE, and RMSE. HrFI (High-resolution Flowering Index) and MYI (Modified Yellowness Index) were able to accurately identify flowering pixels with the least amount of error pixel. The yield simulation model resulted in a pseudo-R2 value of 0.11 for the tested model and, a correlation of 0.91 for validation with RMSE and MAE of 343.1 and 265.3, respectively. Our results indicate that the HrFI index is a better indicator of yield potential compared to NDYI as the metric is well capable of handling within canopy shadows. Further studies are necessary to evaluate the performance of HrFI for medium resolution-UAV and satellite imagery.

Preharvest seed composition estimation using satellite imaging provides critical information for food security planning and management at a regional scale. As one of the staple crops, soybean plays important role in U.S. economic development. Estimating soybean seed composition is the precondition for improving seed quality and meal content at scale, therefore, maintaining U.S. soy competence in international markets. Traditionally, soybean seed compositions are measured after harvest via wet chemistry analysis, which is time-consuming and expensive. This study presents very first, to the best of our knowledge, satellite remote sensing of soybean seed composition. We demonstrate that WorldView-3 satellite imagery and machine/deep learning is powerful tool to predict seed composition from standing crops.

Stomata, microscopic pores on leaf surfaces, regulate the uptake of carbon dioxide and the simultaneous loss of water vapor by leaves. New image acquisition and analysis methods are allowing high-throughput phenotyping of stomatal patterning, which in turn have been applied to better understand the genetic basis of variation in certain species. However, it takes considerable data and effort to train the models, and their ability to accurately detect epidermal structures is constrained to morphologies found within the training data. This issue of context dependency, the inability to perform effectively in novel contexts, is the main hurdle preventing widespread adoption of machine learning in high-throughput phenotyping of intraspecific, interspecific, and environmental variation. Here we show the limited ability of a Mask-RCNN tool, which was previously trained and successfully applied to Zea mays, to analyze images from a closely related grass, Setaria viridis. We then demonstrate successful retraining of the tool to cope with the novel diversity presented by this new species. The stomatal complexes in optical tomography images of mature Setaria leaves were accurately identified by comparison to expert raters (R2 = 0.84). This study highlights the challenge of context dependency for widespread application of machine learning tools for phenotyping plant traits, even in closely related species. At the same time, it also provides a new tool that can be applied to leverage Setaria as a model C4 species, while also providing a roadmap for translation of a machine learning to analyze stomatal patterning in new plant species.

Recent Innovations in Precision Agriculture (PA) are driven by Computer Vision and Data Processing systems to quantify plant parameters. Quantitative analysis of Plant Phenotyping in PA and monitoring morphological traits is a protracting process, precluding the objective and phenotyping pipeline. Greenhouses growing Genetically Modified (GM) crops need to be maintained at constant environmental and simulated conditions. Multiple parameters have to be controlled and regulated inside a greenhouse for effective growth of crops and yield maximisation. Not at all times are these factors derived and so, yield maximisation in greenhouse is an experimental approach to new varieties. For deduced environmental parameters and conditions for certain crops, few other biotic and abiotic factors can hinder or affect growth in certain ways that are not always factored in during calculating parameters conducive for plant growth. Such factors may not always be affecting parametral calculations, but transpose visual cues on plant growth environment such as spectral change in soil values, or minute changes like leaf reflectance or visible changes in plant stimuli to biotic factors. Plant growth is inclusive of multiple environmental variables, and yield maximisation approaches are experimental to finding the optimum derived value for these variables. Computer Vision provides a catalytic approach to predicting optimum parameters for yield maximization in phenomics. Computer Vision and Generative Adversarial Networks (GAN)’s offer a catalytic approach to the time-consuming process, providing a solution to the phenotyping bottleneck. This research proposes a concept of curating data of plant growth over time to predict conditional growth and responsive stimuli of the plant under different situations and how this can affect crop yield. The method proposed here is a non-invasive approach to the existing destructive biomass estimation methods and Frameworks. This methodology of the research focuses on utilizing image parameters modelled using a time series Progressively Growing Generative Adversarial Networks PGGAN to map plant growth patterns and progressive variance in biomass of plant in the Spatio-Temporal Domain. These Generative networks evaluate and predict based on merely raw pixel input excluding dependence on further constraints, feature vectors or parameters influencing data.

Physiological dynamics at plant level are essential but also challenging for precision agriculture applications linked to plant phenotyping. In this study, we explore not only the spatial dynamics of corn in field conditions but also their temporal analysis via skeleton reconstruction of individual plants as a shape descriptor. For this purpose, an optimized approach for high-throughput was developed by point cloud data derived from UAS imagery. The curve-skeleton extraction is calculated based on a constrained Laplacian smoothing algorithm. The experimental setup was performed at the Indiana Corn and Soybean Innovation Center at the Agronomy Center for Research and Education (ACRE) in West Lafayette, Indiana, USA. On July 27th and August 3rd of 2021, two flights were performed over a trial with more than 200 maize plants using a custom designed UAS platform with a Sony Alpha ILCE-7R photogrammetric sensor. RGB images were processed by a standard photogrammetric pipeline by Structure from Motion (SfM) to get a scaled 3D point cloud of the individual corn. Filtering techniques and labeling algorithms were joined together to reconstruct a robust and accurate skeleton of individual maize. Therefore, significant traits such as number, length, growth angle and elongation rate of leaves and stem can be easily extracted. Height variations computed from the skeleton at the two dates show a coefficient of correlation with on-field measurements better than 92%. Our experimental outcomes demonstrate the UAS-data’s ability to provide practical information to efficiently select phenotypes in plant breeding programs.

Continuous collection and analysis of high-resolution phenotype data is critical to develop crops resilient to the consequences of climate change. Though web-accessible tools for parallel, reproducible scientiSic workSlows render big data increasingly tractable, software for plant science remains inadequate for large-scale precision agriculture. Cyberinfrastructure must present minimal barriers to entry, accommodate rapidly changing dependencies, support a wide variety of use cases, and weave together sensors at the edge, laptops, clusters, and cloud storage into a coherent virtual workspace. PlantIT is a web portal intended as such an environment. Platforms like PlantIT and its precursor DIRT [1] permit efSicient phenotyping and equip geographically distributed researchers with a code-optional interface. WorkSlows are published in Docker images, deployed as Singularity containers to public or private computing resources, and monitored in real time. Data are stored automatically in the CyVerse Data Store and can be annotated according to the MIAPPE [2] standard. GitHub integration provides versioning and repositories can be activated with a single conSiguration Sile, like Travis or GitHub Actions. Containers allow for a range of use cases, including image-based trait measurements, 3D reconstructions, morphological growth simulations, and crop modeling. Pseudo-batch/stream processing is also necessary; as data scales, manual batch jobs rapidly become infeasible, and (re-)analysis must occur upon arrival in near-real-time. We suggest web-accessible phenotyping automation software may address bottlenecks and help reveal undiscovered relationships between genes, traits, and the environment.

In an increasingly digitised and data-driven world, there is a pressing need for globally reproducible high-throughput morphological phenotyping, which provides quantitative and objective data and markers of seed quality to guide analysis and research. Current lab-based methods for morphological phenotyping of germinating seeds still largely rely on visual or 2D-imaging technologies with their respective limitations. Here we present the phenoTest, a novel high-throughput 3D-phenotyping technology that alleviates many of the drawbacks of conventional testing and research methods. Using Xray, 3D-volume image data of individual seedlings grown under highly-standardized conditions are captured. Through an AI-based algorithm, all plant organs can be automatically segmented and measured in 3D, currently outputting 50 seedling datasets per 2 minutes. Individual seedlings can be traced over time across the developmental process without disrupting the germination containers. The process can be run in fully-automated, 24h operation and is industrially validated for multiple years. The phenoTest is universally applicable with customized algorithms for all plants and crops, covering the entire range from fine grasses, vegetables to forestry seeds. The process enables a quantitative, objective and reproducible assessment of morphological seedling traits in 4D which can substitute visual gemination and vigor testing and can be harnessed to optimise processing and breeding. We will share data on the effects of a multitude of factors such as seed treatment, ageing, storage and packaging on the speed and quality of seedling development, and the homogeneity, degree of abnormalities, germination capacity and vigor of seed lots of different crop types.

Plant architecture is an important contributing factor for enhanced yield production and quality. The architecture traits are analyzed for crop health monitoring and genetic manipulation for generating high yielding varieties. Computer vision methods applied on 3D pointcloud allow more accurate extraction of architecture traits but consume more time and memory compared to 2D images. This study aims to design light weight 3D deep network for Cotton plant part segmentation and derive seven architectural traits of mainstem height, mainstem diameter, branch inclination angle, branch diameter, and number of branches, nodes, and cotton bolls. The pointcloud data is collected using FARO LiDAR scanner. The mainstem, branches and cotton bolls are manually annotated using Open3D. The preprocessing steps of denoising, normalization and down sampling are applied. 3D Deep network is designed to sample 1024, 512 and 256 points where neighborhood aggregation is performed at radius levels of 1cm, 5cm, and 30cm respectively. Features for remaining points are interpolated. The features from each radius level are concatenated and passed to multi-layer perceptron for pointwise classification. Results indicate that mean IoU and accuracy of 84% and 94% are achieved respectively. A 6.5 times speedup in inference time and 2.4 times reduction in memory consumption compared to Pointnet++ is gained. After applying postprocessing on part segments, an R square value of more than 0.8 and mean absolute percentage error of less than 11% are achieved on all derived architecture traits. The trait extraction results indicate potential utility of this process in plant physiology and breeding programs.

Nitrous oxide (N20) is a greenhouse gas that is three hundred times more potent than carbon dioxide. The majority of N20 emissions worldwide are the result of excess soil nitrogen being metabolized by microbes. It has been hypothesized that crops with better nitrogen uptake efficiency and more roots will reduce excess soil nitrogen therefore reducing N20 emissions. To test this hypothesis, a pilot study was performed in 2021 in collaboration with Iowa State University in which root growth dynamics were captured using RootTracker ™ technology in four commercial maize hybrids. This preliminary study showed a correlation between increased root growth and reduced N20 emissions. Further, we find genetic differences in root growth that is consistent across reps, suggesting that i) cultivar choice impacts N2O emissions and ii) that it is possible to breed for root system architecture to limit N2O emissions. It was also observed that the hybrid with the fastest rate of root growth (lowest N20 emissions) did not reach the greatest soil depth, suggesting early root establishment could be pivotal to more efficient nitrogen uptake. These preliminary results suggest there are differences in root growth by variety that could be exploited to reduce agricultural N20 emissions at scale.

Hyperspectral based prediction of nutrient content in maize leaves Hyperspectral imaging is a promising method to predict crop traits in a high-throughput manner and unlock quantitative genetic studies. A single hyperspectral image can be used to predict several unrelated traits at once using spectral data from 350nm - 2500nm. Researchers have successfully modeled different physiological traits in maize such as vegetative Nitrogen content but the effect of different development stages, genotypes, and treatments on modelling power remains unclear. Here, I explore the ability to model leaf macro- and micro- nutrient content and leaf water content from hyperspectral transmittance data collected with a LeafSpec imaging device. I will compare three different machine learning algorithms; Partial Least Squares Regression, Random Forest and a Convolutional Neural Net to model nutrient content collected from twenty hybrids throughout the 2020 field season in fertilized and not fertilized blocks. Genotypes and development stages excluded from model training are used to externally validate models. Sulfur, Nitrogen, Calcium, Copper, and Iron leaf concentrations were the most amenable nutrients to prediction with coefficient of determination scores from 0.78 - 0.73, respectively. Models trained on samples from a collection of time points were able to accurately predict new time points and genotypes. The findings demonstrate the ability to predict nutrient content in field grown maize over a variety of developmental stages, genotypes, and treatments from a handheld hyperspectral imaging device.

Deep learning (DL) methods have transformed the way we extract plant traits – both under laboratory as well as field conditions. Evidence suggests that “well-trained” DL models can significantly simplify and accelerate trait extraction as well as expand the suite of extractable traits. Training a DL model typically requires the availability of copious amounts of annotated data; however, creating large-scale annotated dataset requires non-trivial efforts, time, and resources. This has become a major bottleneck in deploying DL tools in practice. Self-supervised learning (SSL) methods give exciting solution to this problem, as these methods use unlabeled data to produce pretrained models for subsequent fine-tuning on labeled data, and have demonstrated superior transfer learning performance on down-stream classification tasks. We investigated the application of SSL methods for plant stress classification using few labels. Plant stress classification is a fundamentally challenging problem in that (1) disease classification may depend on abnormalities in a small number of pixels, (2) high data imbalance across different classes, and (3) there are fewer annotated and available plant stress images than in other domains. We compared four different types of SSL methods on two different plant stress datasets. We report that pre-training on unlabeled plant stress images significantly outperforms transfer learning methods using random initialization for plant stress classification. In summary, SSL based model initialization and data curation improves annotation efficiency for plant stress classification tasks.

Cosegmentation is a recent and rapidly emerging and rapidly growing extension of segmentation, which aims to detect the common object(s) in a group of images. Current cosegmentation methods are ideal and effective only for certain dataset types with limited features that still produce errors making it difficult to obtain detailed metrics of object parts. We propose to build a unified, trainable framework that incorporates multiple features of a high-throughput dataset’s segmented images from multiple algorithms using cosegmentation. Specifically, we propose a novel Cosegmentation for Plant Phenotyping Network (CoPPNet) that utilizes a Fully Convolutional Neural Network with a K-Means Clustering feedback loop for optimal temporal loss. The results from this study will set the benchmark for a novel advancement in computer vision segmentation accuracy and plant phenomics to better understand a plant’s environmental interactions for maximal resilience and yield.

Mung bean [Vigna radiata (L.) Wilczek] is a drought-tolerant, short-duration crop, and a rich source of protein and other valuable minerals, vitamins, and antioxidants. The main objectives of this research were (1) to study the root traits related with the phenotypic and genetic diversity of 375 mung bean genotypes of the Iowa (IA) diversity panel and (2) to conduct genome-wide association studies of root-related traits using the Automated Root Image Analysis (ARIA) software. We collected over 9,000 digital images at three-time points (days 12, 15, and 18 after germination). A broad sense heritability for days 15 (0.22–0.73) and 18 (0.23–0.87) was higher than that for day 12 (0.24–0.51). We also reported root ideotype classification, i.e., PI425425 (India), PI425045 (Philippines), PI425551 (Korea), PI264686 (Philippines), and PI425085 (Sri Lanka) that emerged as the top five in the topsoil foraging category, while PI425594 (unknown origin), PI425599 (Thailand), PI425610 (Afghanistan), PI425485 (India), and AVMU0201 (Taiwan) were top five in the drought-tolerant and nutrient uptake “steep, cheap, and deep” ideotype. We identified promising genotypes that can help diversify the gene pool of mung bean breeding stocks and will be useful for further field testing. Using association studies, we identified markers showing significant associations with the lateral root angle (LRA) on chromosomes 2, 6, 7, and 11, length distribution (LED) on chromosome 8, and total root length-growth rate (TRL_GR), volume (VOL), and total dry weight (TDW) on chromosomes 3 and 5. We discussed genes that are potential candidates from these regions. We reported beta-galactosidase 3 associated with the LRA, which has previously been implicated in the adventitious root development via transcriptomic studies in mung bean. Results from this work on the phenotypic characterization, root-based ideotype categories, and significant molecular markers associated with important traits will be useful for the marker-assisted selection and mung bean improvement through breeding.

Crop leaf angle is a crucial feature of plant architecture which influences photosynthetic efficiency and yield. Therefore high-throughput mapping of leaf angle is of extreme interest for both precision agriculture and crop breeding operations. In this study, we propose a UAV-based hybrid approach by combining a radiative transfer model (PROSAIL) and deep neural networks to estimate leaf angle from multi-angular hyperspectral and LiDAR data. PROSAIL can simulate canopy hyperspectral reflectance from a given list of parameters, where Average Leaf Inclination Angle (ALIA) is one of the canopy parameters. The goal is to develop a deep learning-based inversion function that takes UAV hyperspectral reflectance and other PROSAIL parameters as input and estimate ALIA of Maize as output. The other PROSAIL parameters will be estimated using several machine learning pipelines from UAV hyperspectral and LiDAR information. We also propose a multi-angular reflectance scheme where each image pixel will generate multiple simulated and observed reflectance from the overlapping regions using different angles (i.e., solar zenith angle, viewing zenith angle, and relative azimuth angle between the sun and sensor) involved in the PROSAIL simulation. An automated Python-based tool was developed that can calculate all three PROSAIL angles for a given hyperspectral data cube and generate the simulated reflectance for every vegetation pixels per experimental plot. Since the proposed method incorporates both crop information (i.e., PROSAIL) and a data-driven approach (i.e., deep learning), the method can be easily transferable for other study areas and crops, and it can rely on less ground truth data.