CoChat

The DOI matches its Crossref record; the references it cites are not checked yet.

Maize Yield Prediction at an Early Developmental Stage Using Multispectral Images and Genotype Data for Preliminary Hybrid Selection

Open access
Danilevicz, Monica F., Bayer, Philipp E., Boussaid, Farid, Bennamoun, Mohammed, Edwards, David · 2021 · Remote Sensing

Abstract

Assessing crop production in the field often requires breeders to wait until the end of the season to collect yield-related measurements, limiting the pace of the breeding cycle. Early prediction of crop performance can reduce this constraint by allowing breeders more time to focus on the highest-performing varieties. Here, we present a multimodal deep learning model for predicting the performance of maize (Zea mays) at an early developmental stage, offering the potential to accelerate crop breeding. We employed multispectral images and eight vegetation indices, collected by an uncrewed aerial vehicle approximately 60 days after sowing, over three consecutive growing cycles (2017, 2018 and 2019). The multimodal deep learning approach was used to integrate field management and genotype information with the multispectral data, providing context to the conditions that the plants experienced during the trial. Model performance was assessed using holdout data, in which the model accurately predicted the yield (RMSE 1.07 t/ha, a relative RMSE of 7.60% of 16 t/ha, and R2 score 0.73) and identified the majority of high-yielding varieties, outperforming previously published models for early yield prediction. The inclusion of vegetation indices was important for model performance, with a normalized difference vegetation index and green with normalized difference vegetation index contributing the most to model performance. The model provides a decision support tool, identifying promising lines early in the field trial.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.3390/rs13193976

Cites

Statistics

Figures

Figure 1 Scheme of the data processing steps employed for preparing the tabular and spectral datasets as input for the machine learning models. ( a ) Tabular data processing; categorical feature columns were subjected to the steps listed, they were then normalized, one-hot encoded, or used in embedding layers for testing. ( b ) Multispectral image processing. The images collected by the UAV were used to build the orthomosaic, the individual maize plots were then extracted, the pixel values were normalized and, finally, the plot was cut and stacked in random order. The vegetation indices were calculated as new spectral bands, and are shown in the image in the following order: blue, green, red, red edge, NIR, NDVI, NDVI-RE, NDRE, ENVI, CCCI, GNDVI, GLI and OSAVI.
Figure 1 Scheme of the data processing steps employed for preparing the tabular and spectral datasets as input for the machine learning models. ( a ) Tabular data processing; categorical feature columns were subjected to the steps listed, they were then normalized, one-hot encoded, or used in embedding layers for testing. ( b ) Multispectral image processing. The images collected by the UAV were used to build the orthomosaic, the individual maize plots were then extracted, the pixel values were normalized and, finally, the plot was cut and stacked in random order. The vegetation indices were calculated as new spectral bands, and are shown in the image in the following order: blue, green, red, red edge, NIR, NDVI, NDVI-RE, NDRE, ENVI, CCCI, GNDVI, GLI and OSAVI.CC-BY · From the PDF
Figure 2 Schematic representation of the deep learning model architectures; ˆ y indicates the yield prediction output by the model. In ( a ) and ( b ), some layer blocks are repeated, as indicated by the dotted line. ( a ) Tab-DNN architecture with embedding layers. The tab-DNN is trained based on genotype and crop management data, as indicated in the image. ( b ) The sp-DNN architecture is based on a modified resnet18 with a self-attention layer. The sp-DNN was trained with the spectral images plus vegetation indices bands, calculated per pixel. The self-attention layer is present once in the first block repetition, whereas the average pooling layer is not present in the first block iteration.
Figure 2 Schematic representation of the deep learning model architectures; ˆ y indicates the yield prediction output by the model. In ( a ) and ( b ), some layer blocks are repeated, as indicated by the dotted line. ( a ) Tab-DNN architecture with embedding layers. The tab-DNN is trained based on genotype and crop management data, as indicated in the image. ( b ) The sp-DNN architecture is based on a modified resnet18 with a self-attention layer. The sp-DNN was trained with the spectral images plus vegetation indices bands, calculated per pixel. The self-attention layer is present once in the first block repetition, whereas the average pooling layer is not present in the first block iteration.CC-BY · From the PDF
Figure 3 Scheme showing the multimodal model architecture. The tab-DNN and sp-DNN models output an independent yield prediction value (ˆ y) based on their respective input datasets. The weights from the last layer of the tab-DNN and sp-DNN are combined to feed the fusion module, which outputs the multimodal yield prediction (ˆ y).
Figure 3 Scheme showing the multimodal model architecture. The tab-DNN and sp-DNN models output an independent yield prediction value (ˆ y) based on their respective input datasets. The weights from the last layer of the tab-DNN and sp-DNN are combined to feed the fusion module, which outputs the multimodal yield prediction (ˆ y).CC-BY · From the PDF
Image 4
CC-BY · From the PDF
Figure 5 Machine learning model predictions of hybrid yield on the holdout dataset (2017-2019) based on genotype and crop management data. ( a -c ) show the predicted yield (t/ha) in the y-axis in comparison to the ground-truth yield (t/ha) in the x-axis for each sample in the holdout dataset. ( d -f ) present the model prediction error per treatment, the error value was obtained by subtracting the predicted yield from harvested yield. ( g -i ) show the percentage of hybrid plots classified as high, medium and low yield. The x-axis shows the predicted labels, and the y-axis is the ground-truth label.
Figure 5 Machine learning model predictions of hybrid yield on the holdout dataset (2017-2019) based on genotype and crop management data. ( a -c ) show the predicted yield (t/ha) in the y-axis in comparison to the ground-truth yield (t/ha) in the x-axis for each sample in the holdout dataset. ( d -f ) present the model prediction error per treatment, the error value was obtained by subtracting the predicted yield from harvested yield. ( g -i ) show the percentage of hybrid plots classified as high, medium and low yield. The x-axis shows the predicted labels, and the y-axis is the ground-truth label.CC-BY · From the PDF
Figure 6 Sp-DNN yield predictions for the holdout dataset. ( a ) Comparison of the predicted yield (t/ha) to ground-truth yield (t/ha); ( b ) visualization of the model predictions, colored by treatment; ( c ) distribution of maize hybrids by predicted and ground-truth yield (t/ha); ( d ) prediction error (t/ha) of each treatment colored by year; ( e ) percentage of high, medium and low-yielding plots, classified as such by the model; the ground-truth plot label is indicated in the y -axis and predicted labels in the x -axis.
Figure 6 Sp-DNN yield predictions for the holdout dataset. ( a ) Comparison of the predicted yield (t/ha) to ground-truth yield (t/ha); ( b ) visualization of the model predictions, colored by treatment; ( c ) distribution of maize hybrids by predicted and ground-truth yield (t/ha); ( d ) prediction error (t/ha) of each treatment colored by year; ( e ) percentage of high, medium and low-yielding plots, classified as such by the model; the ground-truth plot label is indicated in the y -axis and predicted labels in the x -axis.CC-BY · From the PDF
Figure 7 Cont .
Figure 7 Cont .CC-BY · From the PDF
Figure 7 Images ( a -d ) shows the multimodal fusion model prediction of grain yield (t/ha) from each module and the weighted prediction. Each module prediction is based on their specific data modality, whereas the weighted prediction is obtained by scaling each module's prediction and summing the predicted yield. Image ( e ) shows the model prediction of Figure 7. Images ( a -d ) shows the multimodal fusion model prediction of grain yield (t/ha) from each module and the weighted prediction. Each module prediction is based on their specific data modality, whereas the weighted prediction is obtained by scaling each module's prediction and summing the predicted yield. Image ( e ) shows the model prediction of each maize plot as high-, medium- and low-yield plots (in the y axis), comparing their ground-truth yield in the x -axis. The values within the square indicate the percentage of each category that were predicted to be high-, medium- and low-yield plots. Image ( f ) shows the optimization error of maize plots that presented a high yield but were predicted as medium (in light purple), and plots with a low yield that were predicted to have medium performance (mustard). The plots are separated by year.
Figure 7 Images ( a -d ) shows the multimodal fusion model prediction of grain yield (t/ha) from each module and the weighted prediction. Each module prediction is based on their specific data modality, whereas the weighted prediction is obtained by scaling each module's prediction and summing the predicted yield. Image ( e ) shows the model prediction of Figure 7. Images ( a -d ) shows the multimodal fusion model prediction of grain yield (t/ha) from each module and the weighted prediction. Each module prediction is based on their specific data modality, whereas the weighted prediction is obtained by scaling each module's prediction and summing the predicted yield. Image ( e ) shows the model prediction of each maize plot as high-, medium- and low-yield plots (in the y axis), comparing their ground-truth yield in the x -axis. The values within the square indicate the percentage of each category that were predicted to be high-, medium- and low-yield plots. Image ( f ) shows the optimization error of maize plots that presented a high yield but were predicted as medium (in light purple), and plots with a low yield that were predicted to have medium performance (mustard). The plots are separated by year.CC-BY · From the PDF
Figure 8 Prediction results from the permutation analysis. Image ( a ) shows the impact of randomization on model performance, the x -axis indicates if all VIs were randomized (MSI or RGB) or if only one VI was randomized (minus VI); the y -axis indicates the relative RMSE obtained. Image ( b ) shows the impact of feature randomization on the correlation between features and grain yield. MSI refers to the original multispectral bands (red, green, blue, near-infrared and red edge).
Figure 8 Prediction results from the permutation analysis. Image ( a ) shows the impact of randomization on model performance, the x -axis indicates if all VIs were randomized (MSI or RGB) or if only one VI was randomized (minus VI); the y -axis indicates the relative RMSE obtained. Image ( b ) shows the impact of feature randomization on the correlation between features and grain yield. MSI refers to the original multispectral bands (red, green, blue, near-infrared and red edge).CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

1 more claim about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

The citation lists are queued

They are fetched later and appear here then.

References 66 · Cited by 75

Research works printed but not linked: 1

  1. Kingma, D.P., and Ba, J. Adam: A Method for Stochastic Optimization. arXiv.

Metadata

Authors
Danilevicz, M. F., Bayer, P. E., Boussaid, F., Bennamoun, M., & Edwards, D.
Affiliations
  • The University of Western Australia (AU)
  • School of Computer Science and Software Engineering, University of Western Australia, Perth, WA 6009, Australia
Corresponding author
Edwards, David
Venue
Remote Sensing
Publisher
MDPI AG
Published
2021-10-04
Volume
13
Issue
19
Pages
3976
ISSN
2072-4292 (ISSN-L)
OpenAlex
W3207463009
Cited by
75
References
66
Citations by year
2026: 10 · 2025: 14 · 2024: 21 · 2023: 20 · 2022: 10
FWCI
6.88
Type
journal article
Language
en
Open access
Locations
  • Remote Sensing (publishedVersion, cc-by)
  • UWA Profiles and Research Repository (UWA) (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Australian Research Council (DP210100296, DP200100762)
  • Grains Research and Development Corporation (9177539, 9177591)
Topics
  • Physical Sciences › Environmental Science › Ecology › Remote Sensing in Agriculture
  • Life Sciences › Agricultural and Biological Sciences › Plant Science › Smart Agriculture and AI
  • Physical Sciences › Chemistry › Analytical Chemistry › Spectroscopy and Chemometric Analyses
Keywords
maize yield prediction, early yield prediction, hybrid selection, multispectral imagery, unmanned aerial vehicles, vegetation indices, NDVI, NDREI, genotype data, field management, multimodal deep learning, crop breeding, decision support system

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Maize Yield Prediction at an Early Developmental Stage Using Multispectral Images and Genotype Data for Preliminary Hybrid Selection. https://app.cochat.ai/papers/10.3390/rs13193976
The paper
Danilevicz, M. F., Bayer, P. E., Boussaid, F., Bennamoun, M., & Edwards, D. (2021). Maize Yield Prediction at an Early Developmental Stage Using Multispectral Images and Genotype Data for Preliminary Hybrid Selection. Remote Sensing. https://doi.org/10.3390/rs13193976

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.