Search

In both plant breeding and crop management, interpretability plays a crucial role in instilling trust in AI-driven approaches and enabling the provision of actionable insights. The primary objective of this research is to explore and evaluate the potential contributions of deep learning network architectures that employ stacked LSTM for end-of-season maize grain yield prediction. A secondary aim is to expand the capabilities of these networks by adapting them to better accommodate and leverage the multi-modality properties of remote sensing data. In this study, a multi-modal deep learning architecture that assimilates inputs from heterogeneous data streams, including high-resolution hyperspectral imagery, LiDAR point clouds, and environmental data, is proposed to forecast maize crop yields. The architecture includes attention mechanisms that assign varying levels of importance to different modalities and temporal features that, reflect the dynamics of plant growth and environmental interactions. The interpretability of the attention weights is investigated in multi-modal networks that seek to both improve predictions and attribute crop yield outcomes to genetic and environmental variables. This approach also contributes to increased interpretability of the model's predictions. The temporal attention weight distributions highlighted relevant factors and critical growth stages that contribute to the predictions. The results of this study affirm that the attention weights are consistent with recognized biological growth stages, thereby substantiating the network's capability to learn biologically interpretable features. Accuracies of the model's predictions of yield ranged from 0.82-0.93 R2ref in this genetics-focused study, further highlighting the potential of attention-based models. Further, this research facilitates understanding of how multi-modality remote sensing aligns with the physiological stages of maize. The proposed architecture shows promise in improving predictions and offering interpretable insights into the factors affecting maize crop yields, while demonstrating the impact of data collection by different modalities through the growing season. By identifying relevant factors and critical growth stages, the model's attention weights provide valuable information that can be used in both plant breeding and crop management. The consistency of attention weights with biological growth stages reinforces the potential of deep learning networks in agricultural applications, particularly in leveraging remote sensing data for yield prediction. To the best of our knowledge, this is the first study that investigates the use of hyperspectral and LiDAR UAV time series data for explaining/interpreting plant growth stages within deep learning networks and forecasting plot-level maize grain yield using late fusion modalities with attention mechanisms.

Abstract Hyperspectral data are gaining popularity in remote sensing and signal processing communities because of the increased spectral information relative to multispectral data. Several airborne and spaceborne hyperspectral datasets are publicly available, facilitating the development of various applications and algorithms. However, hyperspectral data are usually limited by their narrow, highly correlated and contiguous spectral bands in both processing and analysis. Moreover, the resolution of available hyperspectral datasets is not sufficiently high for the identification of small objects. Nevertheless, with the rapidly advancing technology, hyperspectral imaging systems can now be mounted on small aerial vehicles for detecting small objects at low altitude. To properly handle these high spectral and spatial resolution data, new or redesigned data processing or analysis pipelines must be developed, but such datasets are limited. In this study, we describe two hyperspectral datasets acquired by a drone and evaluate their radiometric and geometric quality. Based on appropriate data acquisition and processing approaches, our datasets are expected to be useful as testbeds for new algorithms and applications.

Remote sensing enables the rapid assessment of many traits that provide valuable information to plant breeders throughout the growing season to improve genetic gain. These traits are often extracted from remote sensing data on a row segment (rows within a plot) basis enabling the quantitative assessment of any row-wise subset of plants in a plot, rather than a few individual representative plants, as is commonly done in field-based phenotyping. Nevertheless, which rows to include in analysis is still a matter of debate. The objective of this experiment was to evaluate row selection and plot trimming in field trials conducted using four-row plots with remote sensing traits extracted from RGB (red-green-blue), LiDAR (light detection and ranging), and VNIR (visible near infrared) hyperspectral data. Uncrewed aerial vehicle flights were conducted throughout the growing seasons of 2018 to 2021 with data collected on three years of a sorghum experiment and two years of a maize experiment. Traits were extracted from each plot based on all four row segments (RS) (RS1234), inner rows (RS23), outer rows (RS14), and individual rows (RS1, RS2, RS3, and RS4). Plot end trimming of 40 cm was an additional factor tested. Repeatability and predictive modeling of end-season yield were used to evaluate performance of these methodologies. Plot trimming was never shown to result in significantly different outcomes from non-trimmed plots. Significant differences were often observed based on differences in row selection. Plots with more row segments were often favorable for increasing repeatability, and excluding outer rows improved predictive modeling. These results support long-standing principles of experimental design in agronomy and should be considered in breeding programs that incorporate remote sensing.

Yield for biofuel crops is measured in terms of biomass, so measurements throughout the growing season are crucial in breeding programs, yet traditionally time- and labor-consuming since they involve destructive sampling. Modern remote sensing platforms, such as unmanned aerial vehicles (UAVs), can carry multiple sensors and collect numerous phenotypic traits with efficient, non-invasive field surveys. However, modeling the complex relationships between the observed phenotypic traits and biomass remains a challenging task, as the ground reference data are very limited for each genotype in the breeding experiment. In this study, a Long Short-Term Memory (LSTM) based Recurrent Neural Network (RNN) model is proposed for sorghum biomass prediction. The architecture is designed to exploit the time series remote sensing and weather data, as well as static genotypic information. As a large number of features have been derived from the remote sensing data, feature importance analysis is conducted to identify and remove redundant features. A strategy to extract representative information from high-dimensional genetic markers is proposed. To enhance generalization and minimize the need for ground reference data, transfer learning strategies are proposed for selecting the most informative training samples from the target domain. Consequently, a pre-trained model can be refined with limited training samples. Field experiments were conducted over a sorghum breeding trial planted in multiple years with more than 600 testcross hybrids. The results show that the proposed LSTM-based RNN model can achieve high accuracies for single year prediction. Further, with the proposed transfer learning strategies, a pre-trained model can be refined with limited training samples from the target domain and predict biomass with an accuracy comparable to that from a trained-from-scratch model for both multiple experiments within a given year and across multiple years.

Unmanned aerial vehicles (UAVs) equipped with lightweight sensors, such as RGB cameras and LiDAR, have significant potential in precision agriculture, including object detection. Tassel detection in maize is an essential trait given its relevance as the beginning of the reproductive stage of growth and development of the plants. However, compared with general object detection, tassel detection based on RGB imagery acquired by UAVs is more challenging due to the small size, time-dependent variable shape, and complexity of the objects of interest. A novel algorithm referred to as YOLOv5-tassel is proposed to detect tassels in UAV-based RGB imagery. A bidirectional feature pyramid network is adopted for the path-aggregation neck to effectively fuse cross-scale features. The robust attention module of SimAM is introduced to extract the features of interest before each detection head. An additional detection head is also introduced to improve small-size tassel detection based on the original YOLOv5. Annotation is performed with guidance from center points derived from CenterNet to improve the selection of the bounding boxes for tassels. Finally, to address the issue of limited reference data, transfer learning based on the VisDrone dataset is adopted. Testing results for our proposed YOLOv5-tassel method achieved the mAP value of 44.7%, which is better than well-known object detection approaches, such as FCOS, RetinaNet, and YOLOv5.

Leaf area index (LAI) is an important variable for characterizing plant canopy in crop models. It is traditionally defined as the total one-sided leaf area per unit ground area and is estimated by both direct and indirect methods. This paper explores the effectiveness of using light detection and ranging (LiDAR) data to estimate LAI for sorghum and maize with different treatments at multiple times during the growing season from both a wheeled vehicle and Unmanned Aerial Vehicles. Linear and nonlinear regression models are investigated for prediction utilizing statistical and plant structure-based features extracted from the LiDAR point cloud data with ground reference obtained from an in-field plant canopy analyzer (indirect method). Results based on the value of the coefficient of determination (R2) and root mean squared error for predictive models ranged from ∼0.4 in the early season to ∼0.6 for sorghum and ∼0.5 to 0.80 for maize from 40 Days after Sowing to harvest.

Overuse of nitrogen (N), an essential nutrient in food production systems, can lead to health issues and environmental degradation. Two parameters related to N efficiency, N Conversion Efficiency (NCE) and N Internal Efficiency (NIE), measure the amount of total biomass or grain produced, respectively, per unit of N in the plant. Utilizing remote sensing to improve these efficiency measures may positively impact the stewardship of agricultural N use in maize (Zea mays L.) production. We investigated in-season hyperspectral imaging for prediction of end-season whole-plant N concentration (pN), NCE, and NIE, using partial least squares regression (PLSR) models. Image data were collected at two mid-season growth stages (V16/V18 and R1/R2) from manned aircraft and unmanned aerial vehicles for three site years of 5 to 9 maize hybrids grown under 3 N treatments and 2 planting densities. PLSR models resulted in accurate predictions for pN at R6 (R2 = 0.73; R2 = 0.68) and NCE at R6 (R2 = 0.71; R2 = 0.73) from both imaging times. Additionally, the PLSR models based on the R1 images, the second imaging, accurately distinguished the highest and lowest ranked hybrids for pN and NCE across N rates. Neither timepoint resulted in accurate predictions for NIE. Genotype selection efficiency for end-season pN and NCE was increased through the use of the in-season PLSR imaging models, potentially benefiting early breeding screening methods.

Enhancing the nitrogen (N) efficiency of maize hybrids is a common goal of researchers, but involves repeated field and laboratory measurements that are laborious and costly. Hyperspectral remote sensing has recently been investigated for measuring and predicting biomass, N content, and grain yield in maize. We hypothesized that vegetation indices (HSI) obtained mid-season through hyperspectral remote sensing could predict whole-plant biomass per unit of N taken up by plants (i.e., N conversion efficiency: NCE) and grain yield per unit of plant N (i.e., N internal efficiency: NIE). Our objectives were to identify the best mid-season HSI for predicting end-of-season NCE and NIE, rank hybrids by the selected HSI, and evaluate the effect of decreased spatial resolution on the HSI values and hybrid rankings. Analysis of 20 hyperspectral indices from imaging at V16/18 and R1/R2 by manned aircraft and UAVs over three site-years using mixed models showed that two indices, HBSI1 and HBS2, were predictive of NCE, and two indices, HBCI8 and HBCI9, were predictive of NIE for actual data collected from five to nine hybrids at maturity. Statistical differentiation of hybrids in their NCE or NIE performance was possible based on the models with the greatest accuracy obtained for NIE. Lastly, decreasing the spatial resolution changed the HSI values, but an effect on hybrid differentiation was not evident.

High-throughput phenotyping using high spatial, spectral, and temporal resolution remote sensing (RS) data has become a critical part of the plant breeding chain focused on reducing the time and cost of the selection process for the “best” genotypes with respect to the trait(s) of interest. In this paper, the potential of accurate and reliable sorghum biomass prediction using visible and near infrared (VNIR) and short-wave infrared (SWIR) hyperspectral data as well as light detection and ranging (LiDAR) data acquired by sensors mounted on UAV platforms is investigated. Predictive models are developed using classical regression-based machine learning methods for nine experiments conducted during the 2017 and 2018 growing seasons at the Agronomy Center for Research and Education (ACRE) at Purdue University, Indiana, USA. The impact of the regression method, data source, timing of RS and field-based biomass reference data acquisition, and the number of samples on the prediction results are investigated. R2 values for end-of-season biomass ranged from 0.64 to 0.89 for different experiments when features from all the data sources were included. Geometry-based features derived from the LiDAR point cloud to characterize plant structure and chemistry-based features extracted from hyperspectral data provided the most accurate predictions. Evaluation of the impact of the time of data acquisition during the growing season on the prediction results indicated that although the most accurate and reliable predictions of final biomass were achieved using remotely sensed data from mid-season to end-of-season, predictions in mid-season provided adequate results to differentiate between promising varieties for selection. The analysis of variance (ANOVA) of the accuracies of the predictive models showed that both the data source and regression method are important factors for a reliable prediction; however, the data source was more important with 69% significance, versus 28% significance for the regression method.