Detection of apple fruit in an orchard for early yield prediction as a function of crop load

Yield estimation in the commercial production of fruit tree species is becoming an increasingly popular research topic. The outcome, however, is subject to both variability from tree to tree, and from year to year (alternate bearing). Hence, the objective of this contribution was to investigate any potential effects of tree crop load on fruit detection for the prediction of tree yields, as used in image analysis. Approximately 100 apple (Malus domestica Borkh.) trees in Bonn, Germany, were subjected to six crop load modifications, including: (a) mechanical thinning at flowering in April; (b) manual thinning after the June drop; and (c) control without thinning, each time with hand or machine pruning over two years. Trees were photographed three times during fruit ontogeny and these RGB images were converted to YIQ color space and a shape analysis to segment the apple fruits from the background. YCbCr and RGB color spaces were then combined to segment leaves from the tree images. The efficiency of fruit recognition (EFR) improved with fruit ontogeny. EFR was lower without either thinning (38%) or manual pruning (41%) because the lack of either of these measures resulted in many very small fruits. EFR was improved from the unthinned control trees, to those with a single crop load regulation from thinning (hand or mechanical: 49%), to those with the combination of the two crop load regulation practices (58%). Overall, the results indicate that crop load may have a strong influence on fruit recognition in yield prediction work. Apple is used here as a model crop, but the results may be transferable to the wide range of trees where yield estimation is used.