Estimation of satsuma mandarin fruit yield using a drone and hyperspectral sensor

The hyperspectral imaging technology introduced in this study not only significantly enhances the accuracy of citrus yield prediction but also provides crucial data for determining the optimal harvesting time and managing pests and diseases. Recent advancements in drone technology and hyperspectral sensors have greatly expanded their applicability in the agricultural sector. This research utilizes these cutting-edge technologies to propose a new methodology for improving yield prediction accuracy in citrus orchards. In this study, we estimated the yield of Satsuma mandarins (Citrus unshiu Markovich and Citrus reticulata Blanco) using a drone and hyperspectral sensor at three locations in Jeju Island. The collected data were pre-processed (corrected and processed into a mosaic), and dimensions were reduced using a minimum noise fraction. Twenty endmembers were extracted and classified into three groups (background, citrus leaves, and citrus) using the support vector machine (SVM) classification method. The overall accuracy and kappa coefficient, which represent classification accuracy, were 98.87% and 0.83, respectively. Thirty trees were randomly selected from three test areas (Citrus Research Institute, Sinheung-ri, and Odeung-dong test sites), and the pixel values of the extracted citrus fruits and actual weight values of the citrus fruits harvested per tree were compared and analyzed. The corresponding linear regression function was y = y0 + ax; where, a was 0.0555, y0 was 5.7358, and the R-squared value was 0.8099. The yields predicted by substituting the pixel values of citrus fruits into the function were 1,316.20, 12,151.74, and 3,903.56 kg at the Citrus Research Institute, Sinheung-ri, and Odeung-dong test sites, respectively. There was a difference of 5.00–13.30% compared to the actual yield of citrus fruits at each test site.