Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Citrus Fruit Maturity Prediction Utilizing UAV Multispectral Imaging and Machine Learning
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org
Citation: 2021 ASABE Annual International Virtual Meeting 2100495.(doi:10.13031/aim.202100495)
Authors: Israel A Ojo, Lucas Costa, Yiannis Ampatzidis, Fernando Alferez, Sanjay Shukla
Keywords: Maturity prediction, multispectral imaging, supervised machine learning, UAVs
Abstract. Timely harvest of citrus fruit is critical for maximizing yield, fruit quality, and shelf life. The most widely used maturity yardstick for harvesting citrus is the ratio of soluble solids content to titratable acidity of the fruit. Therefore, growers must estimate orchard fruit maturity based on samples that may not represent the entire population and may require harvesting multiple fruits. Additionally, because citrus is climacteric, there is often a tradeoff between a maturity index value that guarantees the best taste quality and a value that provides the desired longevity before final consumption. The objective of this study was to provide a fast, non-destructive method of determining stand fruit quality and predicting optimum maturation for sweet orange (C. sinensis cv Valencia). Machine learning models based on tree canopy spectral reflectance, fruit quality, and harvest date were developed to determine fruit quality. An unmanned aerial vehicle (UAV) was used to obtain spectral data collected from multispectral imaging in order to determine fruit quality parameters. Machine learning algorithms such as support vector machines (SVM), random forest regression (RFR), gradient boosting regression (GBR), and partial least squares regression (PLS) were compared in their ability to make predictions and determine fruit maturation. The mean absolute percentage error (MAPE) based on tree-level evaluation of the models ranged from 0.08 – 0.1 for juice content and 0.49 – 0.37 for maturity index.
(Download PDF) (Export to EndNotes)