Digital twins are a promising approach to create a detailed data foundation providing new insights for breeders and farmers. Using digital twins can make harvest predictions more precise or help detect anomalies (like e.g. diseases or effects of stress) early on. Trees and orchards in particular could benefit from accurate plant data, including plant morphology, objective data like crop size/count per tree, etc. Using a drone-mounted aerial camera we automatically navigate a cherry orchard and generate a series of 2D images as the basis for further processing. Since the processing and amount of data that can be stored on-board a drone is very limited, an approach was chosen to stream the data to the processing computer utilizing a 5G campus network. 3D color point clouds are generated from the 2D images and using AI-based segmentation and image processing each tree is separated into organs, and the tree's morphology is deconstructed. The different plant organs (trunk/branch, leaf, cherry) are then individually digitally modelled, directly generating interpretable values like volume or surface area for each individual plant organ.

Advances in phenotyping tools, genomic methodologies, and analytics strategies provide new tools to assess germplasm merit; however, more work is needed to integrate these systems into modern plant breeding approaches. The objective of this work is to integrate genomics and phenomics for yield prediction in maize. A panel of 830 temperate and tropical inbred lines were evaluated for their testcross performance in 2018, and a subset of 400 testcross hybrids were evaluated in 2021 and 2022. These experiments were performed in West Lafayette, IN in a randomized complete block design with two replications. Remote sensing data was collected on a near weekly basis throughout each growing season for RGB (red-green-blue), LiDAR (light detection and ranging), and VNIR (visible near infrared) hyperspectral data and grain yield was harvested with a plot combine. Remote sensing traits extracted include canopy cover, plot volume, plant height, and NDVI. A GBLUP genomic prediction model was used to estimate yield performance in 2018 using data collected in 2021 and 2022. Remote sensing traits were estimated at regular intervals throughout each growing season using random regression modelling. Grain yield was estimated using the genomic estimated yield and the remote sensing traits in a machine learning model. Preliminary results indicate remote sensing can improve prediction accuracy of grain yield compared to genomic prediction alone even with data only collected before flowering. Improved prediction accuracy could benefit hybrid selection, increase genetic gain, and reduce cost in a breeding program.

Critical factors that determine crop yields are located underground, making them difficult to analyze. Traditionally, these factors have been measured by growing plants in clear media and measuring traits with visible imaging. Modern phenomics technologies use one or several imaging modalities to capture traits that reflect plant physiology or performance. Analytical techniques for plant phenomics are a crucial part of approaches to achieving desirable agronomic and biological traits. Advances in sensor technologies have paved the way for faster and more efficient plant phenotyping, with methods adapted from disciplines like high-resolution 3D X-Ray computed tomography (CT). A crucial step in their analysis is segmentation-the identification and classification of the scan's voxels as "root" or "non-root". Unlike roots in transparent mediums, roots in non-transparent mediums are difficult to segment from their surrounding materials as root and non-root voxels have overlapping CT values. The challenge we address is the development of neural-driven approaches for volumetric semantic segmentation of plant roots in 3D CT scans, and discuss subsequent trait extraction methods that enable the quantification of root systems and their traits in several agriculturally

Grain and seed properties can be evaluated using near-infrared spectroscopy and other methods for post-harvest quality assessment. Hyperspectral imaging combines spectroscopy with spatial information, which provides additional features that may improve predictive models of seed traits. To assess the ability of deep learning models to use hyperspectral data for predicting phenotypes, we first aimed to predict the genotype of maize seeds. Previous work achieved high identification accuracy between a small set of genotypes using either RGB images or hyperspectral data, and we hypothesized that high spectral resolution (350-1000nm) hyperspectral data would outperform simple RGB data in our study. Our dataset consisted of hyperspectral images of maize seeds from 47 inbred lines, including the 26 NAM lines, with 96 individual seeds per genotype. We evaluated the difference in genotype identification accuracy using three different representations of the individual seed data: 1) using the whole scan, containing the reflectance at 580 different wavelengths, 2) using a subset containing the reflectance at 3 different wavelengths corresponding to a pseudo-RGB image, and 3) a gray-scale image derived from the pseudo-RGB image. We fine-tuned VGG11, a popular convolutional neural network, using 85% of the individual seed data for each of the representations. We obtained around 90% genotype prediction accuracy on the unseen data for both the whole scan and the pseudo-RGB data, and 72% genotype prediction accuracy using the gray-scale data. The results indicate that the shape and color information contained in RGB images might be sufficient for the task of maize seed genotype identification.

High night air temperature stress (HNT) challenges rice production. Findings indicate 10% yield reduction for every 1 o C of increase in night air temperature. The responses of rice to HNT stress have been analyzed in limited number of genotypes mostly under greenhouse conditions. One of the limits for these studies under field conditions is implementing HNT stress on critical rice growth stage. The physiological and metabolic responses of rice to HNT stress under field conditions are not fully understood, thus, field studies are needed. Field-based phenotyping infrastructure that can house rice germplasm and stress imposition using computer-based system basing on ambient temperature still do not exist. In this study, six high tunnel greenhouses were built in a field experimental station in Harrisburg, AR in a split-plot design. These movable infrastructures fitted 310 rice accessions from the Rice Diversity Panel 1 (RDP1) and 10 hybrids from RiceTec. Each high tunnel greenhouse had heating and a cyber-physical system that recorded ambient air temperature and increased night air temperature relative to ambient temperature at the flowering stage. The system successfully imposed HNT stress of 4.01 o C and 3.94 o C as recorded by Raspberry Pi sensors for two weeks in the 2019 and 2020 cropping seasons, respectively. These greenhouses were able to endure constant flooding and resist heavy rain and 40-50 miles/h winds. Grain quality and other biochemical assays are still ongoing to fully assess the effects of HNT in the rice accessions and the hybrids.

Pores in the leaf epidermis called stomata allow plants to take up carbon dioxide for photosynthesis, but are also pathways for water vapor loss. New image acquisition and analysis methods are allowing high-throughput phenotyping of stomatal patterning, which can be 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 by 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 trained and successfully applied to Zea mays, to analyze images from a closely related grass called Setaria viridis. We then demonstrate successful retraining of the tool to cope with the novel amounts of 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 (R 2 = 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, and a roadmap for the translation of a machine learning tool to analyze stomatal patterning in diverse datasets of new plant species.

Sara Tirado Tolosa1

A document by Sierra Young. Click on the document to view its contents.

Miscanthus is one of the most promising perennial crops for bioenergy production, with high yield potential and low environmental footprint. The increasing interest in this crop requires accelerated selection and the development of new screening techniques. The development of analytical methods for improved estimation and reduced manual inspection are needed to better characterize the effects of genetics and the environment in key traits under field conditions. We used persistent multispectral and photogrammetric UAV time-series imagery collected 10 times in the season, together with ground-truth data over thousands miscanthus accessions to determine flowering time, culm height, and biomass yield traits. We compared the performance of Convolutional Neural Network (CNN) architectures that used image data from single dates (2D-spatial) versus the integration of multiple dates (3D-spatio-temporal) architectures to evaluate the value of persistent monitoring and the type of features to predict the traits. The ability of UAV-based remote sensing to rapidly and non-destructively assess large-scale genetic variation in flowering time, height and biomass production was improved through use of 3D-spatio-temporal CNN architectures versus 2D-spatial CNN architectures. The performance gains of the best 3D-spatio-temporal analyses compared to the best 2D-spatial architectures manifested in up to: 23 % improvements in R 2 and 20 % reduction in mean absolute error (MAE). The integration of

Stomatal conductance (SC) was utilized to indicate the rate of gas and water exchange through stomata on the leaf surface. When crops are experiencing water stress during the daytime, their stomata will close to prevent water loss. However, the crop will also receive less CO2 for photosynthesis under this condition. Consequently, accurate and efficient SC prediction can improve irrigation efficiency, particularly in the present day when water is scarce. The common way to estimate SC nowadays is to make predictions using other variables that can be measured quickly, such as conventional regression analysis. The limitation of conventional regression analysis is that it does not account for changes in SC when crops are subjected to both prolonged drought and high temperature stress. We intend to use time series prediction techniques, such as Long Short-Term Memory (LSTM), to determine the correlation between variables at different time scales. The objective of this study is to (1) Investigating the relationship between SC and persistent weather patterns. (2) Predict SC based on continuously collected soil moisture, plant canopy temperature and weather information. In this study, we measured the SC of soybean and maize by using a handheld leaf porometer. This was then compared to short-term historical crop SC predictions based on canopy temperature, soil moisture, and weather conditions. Autoregressive Integrated Moving Average (ARIMA) and LSTM based on Recurrent neural network (RNN) are used to predict SC, and the results are compared with the conventional modeling method. More details will be presented at the NAPPN conference.

Jinjeong Lee, Su Jung Ra, Sung Mi Kim, Eunhee Kim, Myeong-Jin Kang,

Object detection algorithms have heavily relied on deep learning techniques to estimate the count of grapes as the resulting quality of grapes is directly correlated to its yield. With temporal analytics, early actions and logistical organizing can be performed to maintain the quality of grapes. However, the issue with using these object detection algorithms is that they are limited to counting only the visible grapes, thus omitting hidden grapes, and in turn affecting the true estimate of grape yield. Many grapes are occluded because of either the compactness of the bunch cluster, or due to canopy interference. Therefore, models need to be able to estimate the unseen berries in order to give a near true yield estimate. An end-to-end framework is proposed in which the grape clusters are first segmented using a deep learning model, after which the extracted candidate regions of grape clusters are fed to a CNN regression model that can estimate the count of berries by incorporating a correction factor. A new dataset is also proposed which consists imagery of grape clusters, along with their ground truth values of grape count and weight. The proposed framework will also be tested using three open-source datasets and will encourage future research in determining which features of grapes can be leveraged to correct grape counting algorithms and produce higher accurate results.

Resource allocation drives the above-ground distribution of mass in grass plants across discrete developmental units called phytomers. Although the number of phytomers varies in genetically-identical grasses, there frequently isn't an associated variance in some summary phenotypes. To understand what may be driving this, we tracked the growth of 30 S. italica plants from genotypes B100 and A10.1. We experimentally observed that plants from the genotype B100 had between 20 and 22 phytomers, while plants from the genotype A10.1 had between 7 and 9 phytomers. B100 plants with more phytomers (e.g., 22) did not grow taller or have more total leaf length, despite having more leaves than plants with fewer phytomers (e.g., 20). A10.1 plants with more phytomers (e.g., 9) did grow taller and had more total leaf length than those with fewer phytomers (e.g., 7). We developed a dynamical model to determine if these patterns are emergent from the underlying growth structure. The model is parameterized using the number of phytomers and related developmental time parameters: leaf emergence, stem and leaf elongation time, panicle emergence, and flowering time. The model uses the semi-sequential nature of phytomer growth as its structure. The model predicts that differences in timing of the shift to reproductive growth could explain the patterns observed. Experimental measurements suggest this shift is primarily due to tuning the developmental time parameter controlling the units contributing to the stem, rather than leaves.

Maize (Zea mays ssp. Mays) is one of the most essential cereal crops in the world. As climate changes, breeding for high nitrogen-use efficiency maize genetic materials without sacrificing the yield becomes more urgent than anytime before. Image-based high-throughput phenotyping, functioning as a key element in plant breeding efforts, is critically important and has the potential to relieve difficulties in phenotypic scoring on breeding pipelines. Numberless studies using RGB images related to biomass and agronomically important traits have mostly focused on above-ground traits. The belowground root-related traits, however, have not been intensively studied. The objective of this study is to investigate the root architecture and phenotypic properties (number of aerial roots, stem diameter, internode length, and fresh root weight) of roots of the hybrid and inbred maize lines grown under low and high nitrogen conditions. In this study, a collection of BGEM lines (n = 304 inbred and n = 197 hybrids) were planted on the field in low and high nitrogen conditions. The root samples of n = 2,100 plants were collected, and the soil around the roots was washed out for automated image-based phenotyping. The roots were imaged with the completely automated conveyor belt LemnaTec system. A high variation in root structure, stem diameter, number of aerial roots, and internode length was observed among the genotypes. Our root phenotyping pipeline and traits extracted from these images will enhance root biology and facilitate breeding for below-ground root traits.

Source-Sink Regulated Senescence in Commercial Maize Germplasm

Hyperspectral imaging is a non-destructive imaging technique used in plant phenotyping to collect and analyze an array of electromagnetic information in visible (380-700 nm) and near-infrared wavelengths region (700-2,500 nm). Hyperspectral imaging can provide information of plant responses under various biotic and abiotic stress, e.g., drought, temperature rising, disease, and nutrition deficiency. We present a hyperspectral data processing pipeline designed for the data collected at Ag Alumni Seed Phenotyping Facility (AAPF) in Purdue University, USA. The procedure consists of initializing a processing session, radiometric calibration with white and dark references, geometric calibration (registration) of visible and near infrared (VNIR) and shortwave infrared (SWIR) images, vegetation and non-vegetation classification, vegetation indices calculation of a plant area, exporting data products, and quality control. In concern of large data size of hyperspectral data, we highlight the need to save memory usage during computation and save disk space for data products. We also address the need of human interpretable images in the hyperspectral data products for plant scientists without experiences in hyperspectral imaging. We expect the developed procedure could improve robustness of large hyperspectral data processing and promote the usage of hyperspectral data by increasing interpretability.

A document by Jose C. Tovar. Click on the document to view its contents.

Minirhizotron technique (MR), an image-based root phenotyping technology, has expanded our understanding of in situ root responses to changing environmental conditions. However, the conventional manual approaches to capturing and analyzing images are time-consuming, thus constraining the size and frequency of sampling and interpretation. To address this bottleneck, an automated MR system was developed, which worked in a fully automatic manner according to the pre-set schedule. Users can remotely control the system and download images. This system was tested in a net house by imaging bell pepper roots on a daily basis, which shows superior performance over commercial manual MR systems in terms of image resolution and sampling frequency. Besides, an image analysis model was built on convolutional neural networks to estimate root length from MR images directly without segmentation in the training process. This model was trained on a dataset of ~18,000 tomato root images taken by a manual MR camera and was used for estimating root length on 832 bell pepper root images taken by the automated MR camera. The high correlation coefficient (R 2 = 0.854) between the model estimation and manual measurement proved that this model generalizes well over different crop roots and camera types. Therefore, high-frequency in situ root phenotyping can be achieved by the automated Minirhizotron image acquisition and analysis tools proposed here.