High-Throughput Phenotyping of Soybean Maturity Using Time Series UAV Imagery and Convolutional Neural Networks

High-throughput phenotyping using VIS/NIR spectroscopy in the classification of soybean genotypes for grain yield and industrial traits

Employing visible and near infrared sensors in high-throughput phenotyping provides insight into the relationship between the spectral characteristics of the leaf and the content of grain properties, helping soybean breeders to direct their program towards improving grain traits according to researchers' interests. Our research hypothesis is that the leaf reflectance of soybean genotypes can be directly related to industrial grain traits such as protein and fiber contents. Thus, the objectives of the study were: (i) to classify soybean genotypes according to the grain yield and industrial traits; (ii) to identify the algorithm(s) with the highest accuracy for classifying genotypes using leaf reflectance as model input; (iii) to identify the best input data for the algorithms to improve their performance. A field experiment was carried out in randomized block design with three replications and 32 soybean genotypes. At 60 days after emergence, spectral analysis was carried out on three leaf samples from each plot. A hyperspectral sensor was used to capture reflectance between the wavelengths from 450 to 824 nm. Representative spectral bands were selected and grouped into means. After harvest, grain yield was assessed and laboratory analyses of industrial traits were carried out. Spectral, industrial traits and yield data were subjected to statistical analysis. Data were analyzed by the following machine learning algorithms: J48 (J48) and REPTree (DT) decision trees, Random Forest (RF), Artificial Neural Networks (ANN), Support Vector Machine (SVM), and conventional Logistic Regression (LR) analysis. The clusters formed were used as the output of the models, while two groups of input data were used for the input of the models: the spectral variables (WL) noise-free obtained by the sensor (450-828 nm) and the spectral means of the selected bands (SB) (450.0-720.6 nm). Soybean genotypes were grouped according to their grain yield and industrial traits, in which the SVM and J48 algorithms performed better at classifying them. Using the spectral bands selected in the study improved the classification accuracy of the algorithms.

Multispectral-derived genotypic similarities from budget cameras allow grain yield prediction and genomic selection augmentation in single and multi-environment scenarios in spring wheat

With abundant available genomic data, genomic selection has become routine in many plant breeding programs. Multispectral data captured by UAVs showed potential for grain yield (GY) prediction in many plant species using machine learning; however, the possibilities of utilizing this data to augment genomic prediction models still need to be explored. We collected high-throughput phenotyping (HTP) multispectral data in a genotyped multi-environment large-scale field trial using two cost-effective cameras to fill this gap. We tested back to back the prediction ability of GY prediction models, including genomic (G matrix), multispectral-derived (M matrix), and environmental (E matrix) relationships using best linear unbiased predictor (BLUP) methodology in single and multi-environment scenarios. We discovered that M allows for GY prediction comparable to the G matrix and that models using both G and M matrices show superior accuracies and errors compared with G or M alone, both in single and multi-environment scenarios. We showed that the M matrix is not entirely environment-specific, and the genotypic relationships become more robust with more data capture sessions over the season. We discovered that the optimal time for data capture occurs during grain filling and that camera bands with the highest heritability are important for GY prediction using the M matrix. We showcased that GY prediction can be performed using only an RGB camera, and even a single data capture session can yield valuable data for GY prediction. This study contributes to a better understanding of multispectral data and its relationships. It provides a flexible framework for improving GS protocols without significant investments or software customization.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11032-024-01449-w.

Prediction of Buckwheat Maturity in UAV-RGB Images Based on Recursive Feature Elimination Cross-Validation: A Case Study in Jinzhong, Northern China

Buckwheat is an important minor grain crop with medicinal and edible functions. The accurate judgment of buckwheat maturity is beneficial to reduce harvest losses and improve yield. With the rapid development of unmanned aerial vehicle (UAV) technology, it has been widely used to predict the maturity of agricultural products. This paper proposed a method using recursive feature elimination cross-validation (RFECV) combined with multiple regression models to predict the maturity of buckwheat in UAV-RGB images. The images were captured in the buckwheat experimental field of Shanxi Agricultural University in Jinzhong, Northern China, from September to October in 2021. The variety was sweet buckwheat of "Jinqiao No. 1". In order to deeply mine the feature vectors that highly correlated with the prediction of buckwheat maturity, 22 dimensional features with 5 vegetation indexes, 9 color features, and 8 texture features of buckwheat were selected initially. The RFECV method was adopted to obtain the optimal feature vector dimensions and combinations with six regression models of decision tree regression, linear regression, random forest regression, AdaBoost regression, gradient lifting regression, and extreme random tree regression. The coefficient of determination (R2) and root mean square error (RMSE) were used to analyze the different combinations of the six regression models with different feature spaces. The experimental results show that the single vegetation index performed poorly in the prediction of buckwheat maturity; the prediction result of feature space "5" combined with the gradient lifting regression model performed the best; and the R2 and RMSE were 0.981 and 1.70 respectively. The research results can provide an important theoretical basis for the prediction of the regional maturity of crops.

The field phenotyping platform's next darling: Dicotyledons

The genetic information and functional properties of plants have been further identified with the completion of the whole-genome sequencing of numerous crop species and the rapid development of high-throughput phenotyping technologies, laying a suitable foundation for advanced precision agriculture and enhanced genetic gains. Collecting phenotypic data from dicotyledonous crops in the field has been identified as a key factor in the collection of large-scale phenotypic data of crops. On the one hand, dicotyledonous plants account for 4/5 of all angiosperm species and play a critical role in agriculture. However, their morphology is complex, and an abundance of dicot phenotypic information is available, which is critical for the analysis of high-throughput phenotypic data in the field. As a result, the focus of this paper is on the major advancements in ground-based, air-based, and space-based field phenotyping platforms over the last few decades and the research progress in the high-throughput phenotyping of dicotyledonous field crop plants in terms of morphological indicators, physiological and biochemical indicators, biotic/abiotic stress indicators, and yield indicators. Finally, the future development of dicots in the field is explored from the perspectives of identifying new unified phenotypic criteria, developing a high-performance infrastructure platform, creating a phenotypic big data knowledge map, and merging the data with those of multiomic techniques.

Response of a Diverse European Soybean Collection to "Short Duration" and "Long Duration" Drought Stress

Drought causes significant damage to a high value crop of soybean. Europe has an increasing demand for soybean and its own production is insufficient. Selection and breeding of cultivars adapted to European growth conditions is therefore urgently needed. These new cultivars must have a shorter growing cycle (specifically for adaptation to North-West Europe), high yield potential under European growing conditions, and sufficient drought resistance. We have evaluated the performance of a diverse collection of 359 soybean accessions under drought stress using rain-out shelters for 2 years. The contrasting weather conditions between years and correspondingly the varying plant responses demonstrated that the consequences of drought for an individual accession can vary strongly depending on the characteristics (e.g., duration and intensity) of the drought period. Short duration drought stress, for a period of four to 7 weeks, caused an average reduction of 11% in maximum canopy height (CH), a reduction of 17% in seed number per plant (SN) and a reduction of 16% in seed weight per plant (SW). Long duration drought stress caused an average reduction of 29% in CH, a reduction of 38% in SN and a reduction of 43% in SW. Drought accelerated plant development and caused an earlier cessation of flowering and pod formation. This seemed to help some accessions to better protect the seed yield, under short duration drought stress. Drought resistance for yield-related traits was associated with the maintenance of growth under long duration drought stress. The collection displayed a broad range of variation for canopy wilting and leaf senescence but a very narrow range of variation for crop water stress index (CWSI; derived from canopy temperature data). To the best of our knowledge this is the first study reporting a detailed investigation of the response to drought within a diverse soybean collection relevant for breeding in Europe.

Classification of Soybean Pubescence from Multispectral Aerial Imagery

The accurate determination of soybean pubescence is essential for plant breeding programs and cultivar registration. Currently, soybean pubescence is classified visually, which is a labor-intensive and time-consuming activity. Additionally, the three classes of phenotypes (tawny, light tawny, and gray) may be difficult to visually distinguish, especially the light tawny class where misclassification with tawny frequently occurs. The objectives of this study were to solve both the throughput and accuracy issues in the plant breeding workflow, develop a set of indices for distinguishing pubescence classes, and test a machine learning (ML) classification approach. A principal component analysis (PCA) on hyperspectral soybean plot data identified clusters related to pubescence classes, while a Jeffries-Matusita distance analysis indicated that all bands were important for pubescence class separability. Aerial images from 2018, 2019, and 2020 were analyzed in this study. A 60-plot test (2019) of genotypes with known pubescence was used as reference data, while whole-field images from 2018, 2019, and 2020 were used to examine the broad applicability of the classification methodology. Two indices, a red/blue ratio and blue normalized difference vegetation index (blue NDVI), were effective at differentiating tawny and gray pubescence types in high-resolution imagery. A ML approach using a support vector machine (SVM) radial basis function (RBF) classifier was able to differentiate the gray and tawny types (83.1% accuracy and kappa = 0.740 on a pixel basis) on images where reference training data was present. The tested indices and ML model did not generalize across years to imagery that did not contain the reference training panel, indicating limitations of using aerial imagery for pubescence classification in some environmental conditions. High-throughput classification of gray and tawny pubescence types is possible using aerial imagery, but light tawny soybeans remain difficult to classify and may require training data from each field season.

Global Wheat Head Detection 2021: An Improved Dataset for Benchmarking Wheat Head Detection Methods

The Global Wheat Head Detection (GWHD) dataset was created in 2020 and has assembled 193,634 labelled wheat heads from 4700 RGB images acquired from various acquisition platforms and 7 countries/institutions. With an associated competition hosted in Kaggle, GWHD_2020 has successfully attracted attention from both the computer vision and agricultural science communities. From this first experience, a few avenues for improvements have been identified regarding data size, head diversity, and label reliability. To address these issues, the 2020 dataset has been reexamined, relabeled, and complemented by adding 1722 images from 5 additional countries, allowing for 81,553 additional wheat heads. We now release in 2021 a new version of the Global Wheat Head Detection dataset, which is bigger, more diverse, and less noisy than the GWHD_2020 version.