Detecting and counting citrus fruit on the ground using machine vision

A machine vision system for estimating number of citrus fruit drop was developed in this study. The objectives of this study were to design rugged hardware, to develop an image processing algorithm for accurate estimation of fruit count and to conduct field experiments. Image acquisition hardware was developed to be used in a commercial citrus grove specifically for unfavorable imaging conditions. The image processing algorithm included normalization of intensity, citrus fruit detection by a logistic classifier, and least square circle fitting. Accuracy of the algorithm was analyzed using two different methods. Firstly, the ability of detecting citrus fruit by the algorithm without any missed fruit was analyzed. The accuracy varied within three trials, and the highest was 89.5 percent. The second analysis was for the ability to avoid false positives which represent incorrect detection of the background object as a citrus. The percentage of false positive detection also varied between the trials. The highest error was 16.2 percent and the lowest error was 9.8 percent. Result of the experiments showed that each trial had different number and mass of citrus fruit drop. This was because each area in the images had different site-specific variable factors such as nutrient level, soil pH, disease, canopy size etc. The machine vision algorithm can be modified for more advanced application such as immature citrus fruit drop detection and counting during mechanical harvesting and early yield estimation.