Spectral Sensing of Different Citrus Varieties for Precision Agriculture

Researchers have used different forms of non-destructive computer vision sensing oncitrus for years; however, no system has been developed to identify maturing green citrus fruit whilethey are still on tree. This project is a preliminary study as to the validity of distinguishing greencitrus fruit varieties from leaves using only their spectral characteristics.A spectrophotometer was used to measure diffuse reflectance of green leaves and three citrus fruitvarieties (Orlando Tangelo, Hamlin, and Valencia) in the 200 nm to 2500 nm range. The growingpattern and maturing process of the fruit samples were studied for optimal classification. In addition,moisture contents were calculated and compared with sample spectral characteristics to betterunderstand the role moisture has in determining the fruits spectral characteristics.The best wavelengths for green fruit identification were determined using discriminablity. Thesefeature spaces used in discriminant analysis to distinguish between fruit and leaf were proven highlyaccurate. Using two-thirds of the total data as training data and one-third as validation data, a R2 ashigh as 1.0 was found possible. Calculations using all samples found the optimal wavelengths forleaf/fruit separation were 881, 781, and 1383 nm. These results prove that highly accurateidentification of green citrus fruits from leaves is possible while using diffuse reflectance spectralbands.