Identification and Determination of the Number of Green Citrus Fruit under Different Ambient Light Conditions

Yield mapping by machine aided harvesting requires automatic detection and counting of fruit in a tree canopy. However, occlusion, varying illumination, and similarity with the background make fruit identification a very challenging task. Moreover, green citrus detection within green canopy is a very difficult problem due to the issues previously mentioned. In this study, a novel and simple technique is discussed to detect green citrus in the tree canopy under natural outdoor conditions by utilizing shape and texture information. First a shape analysis is conducted to detect as many as citrus as possible. Texture classification combined with a support vector machine, graph based connected component algorithm and Hough line detection were used to remove the false positives in the first stage. Then keypoints were detected by scale invariant feature transform algorithm and used to remove additional false positives. A majority voting scheme has been implemented to make the algorithm more robust. The algorithm was able to accurately detect and count over 81% of citrus fruit in a validation set of images acquired from a citrus grove.