?? AI for Sugarcane: Phenotyping with UAVs

Phenotyping sugarcane without UAVs faces significant challenges. Given the large scale of sugarcane fields, it's impractical to collect detailed and frequent data manually, leading to incomplete or inaccurate information. Additionally, traditional methods lack the high spatial and temporal resolution provided by UAVs.

Hence, today we will look at plant phenotyping for sugarcane with UAVs and further data processing using AI models.

UAVs equipped with multispectral cameras can quickly cover large fields, capturing high-resolution data that allows for detailed analysis of crop health and growth. The next three case studies explain the practical link between using drones and machine learning to get insights about sugarcane.

Mapping Gaps in Sugarcane Fields with UAV Imagery

Country: ???? Brazil

Published: 18 December 2021

This study evaluates the accuracy of using UAV RGB imagery to map gaps in sugarcane fields, focusing on the impact of flight height and pixel size on gap detection.

The study involved a series of flights using a DJI Phantom 4 UAV equipped with an RGB sensor (DJI Zenmuse X3 camera) to capture images of sugarcane fields at various heights (0.5m, 0.9m, 1.0m, 1.2m, and 1.7m) and pixel sizes (3.5cm, 6.0cm, and 8.2cm).

This study employs artificial intelligence to create orthomosaic maps, identify and segment low-intensity pixels based on chromatic similarity and topological proximity, and distinguish plants from soil. The AI then iteratively identifies gaps, measures their dimensions, and converts this information into vectorial files for further analysis and validation

Key findings indicate that lower flight heights and smaller pixel sizes significantly improve the accuracy of gap detection. For instance, a flight height of 0.5m with a pixel size of 3.5cm resulted in the most precise gap measurements, with an R2 value of 0.95 and a mean absolute error (MAE) of 0.24m.

These results demonstrate the potential of UAV imagery for precise agricultural monitoring.

Agricultural researchers, precision agriculture practitioners, and farm managers can practically apply these findings to enhance the accuracy of crop monitoring and management.

Main tools/technologies

• UAVs (Unmanned Aerial Vehicles)
• RGB cameras
• Image analysis software
• Linear regression analysis
• Chromatic similarity and topological proximity algorithms

Predicting Sugarcane Traits with UAVs

Country: ???? United States

Published: February 2023

This study explores the use of aerial imagery and machine learning to predict morpho-physiological traits in sugarcane, aiming to enhance breeding programs and crop management.

The research employed a DJI Matrice 600 Pro hexacopter equipped with a Pika-L 2.4 hyperspectral imaging sensor to capture images of sugarcane fields at various growth stages. Multiple machine learning algorithms, including gradient boosting regression trees (GBRT), random forest, and support vector regression, were evaluated using a five-fold cross-validation methodology. The study focused on predicting key traits such as:

1. Leaf greenness (indication of leaf chlorophyll)
2. Leaf Area Index (LAI)
3. Normalized Difference Vegetation Index (NDVI)
4. Plant height
5. Number of millable stalks per hectare

Vegetation indices, particularly NDVI, were derived from the imagery and used as input features for the prediction models.

Key findings reveal that the GBRT model outperformed other algorithms in predicting sugarcane traits.

For plant height prediction, the model achieved mean absolute percentage errors (MAPE) ranging from 13.7% to 21.2% across different growth stages and crop cycles. The prediction of millable stalks per hectare was more accurate, with MAPE values between 8.6% and 13.4%. These results demonstrate the potential of combining UAV imagery with machine learning for non-invasive and efficient estimation of important sugarcane traits, which can significantly accelerate breeding programs and improve crop management decisions.

Sugarcane breeders, agronomists, and precision agriculture practitioners can practically apply these findings to enhance crop monitoring, selection processes, and yield prediction.

Main tools/technologies

• Unmanned Aerial Vehicles (UAVs)
• RGB and multispectral sensors
• Machine learning algorithms (GBRT, random forest, SVR)
• Vegetation indices (NDVI)
• Python programming language
• Five-fold cross-validation

Advanced Machine Learning for Sugarcane Yield Prediction

Country: ???? Brazil

Published: 23 August 2022

Using multispectral UAV images and machine learning algorithms, this study aims to accurately predict sugarcane biometric parameters, enhancing yield prediction capabilities.

The researchers utilized UAVs equipped with multispectral camera MicaSense RedEdge-M to capture images of sugarcane fields. They employed machine learning algorithms, specifically Multiple Linear Regression (MLR) and Random Forest (RF), to analyze these images and predict three key biometric parameters: number of tillers (NT), plant height (PH), and stalk diameter (SD). Data was collected from five sugarcane varieties over multiple time points to ensure robustness.

Key findings indicate that RF models outperformed MLR models, with higher R2 values for NT and PH (R2 > 0.70). The spectral bands Blue, Green, and NIR were the most effective predictors. The fusion of images from multiple time points (GDD = 349 and 397) provided the most accurate results, significantly improving sugarcane yield prediction accuracy.

Farmers and agronomists can practically apply these results to improve sugarcane yield predictions and optimize crop management.

Main Tools/Technologies:

1. UAVs with multispectral cameras
2. Machine Learning algorithms: Multiple Linear Regression (MLR) and Random Forest (RF)
3. Spectral analysis of Blue, Green, and NIR bands
4. Software: Agisoft Metashape, QGIS, Python with scikit-learn library

??What's next in sugarcane tech?

In the next edition of "AI for Sugarcane" we would like to delve deeper into the topic of yield prediction. Additionally, we have pest and disease detection as two next directions for our work.

?What would you like to read about?

Share your thoughts in the comment below ??

If you find this newsletter useful, please, support it with your like or share to your colleagues ??

See you tomorrow!

Have a productive UAV flights over your sugarcane crops,

Maryna Kuzmenko , Chief Inspiration Officer at Petiole Pro

#sugarcane

Photo credit for the cover image:

Barbosa Júnior, M.R.; Tedesco, D.; Corrêa, R.d.G.; Moreira, B.R.d.A.; Silva, R.P.d.; Zerbato, C. Mapping Gaps in Sugarcane by UAV RGB Imagery: The Lower and Earlier the Flight, the More Accurate. Agronomy 2021, 11, 2578. https://doi.org/10.3390/agronomy11122578

References in "AI for Sugarcane"