Recently, a machine vision technology has shown its popularity for automating visual inspection. Many studies proved that the machine vision system can successfully estimate external qualities of fruit as good as manual inspection. However, introducing mechanical harvesters to citrus industry caused the following year’s yield loss due to the loss of immature young citrus. In this study, a machine vision system on a conveyor belt was developed to inspect mechanically harvested citrus fruit. Object based classification was conducted on RGB images acquired on the conveyor belt. Three ensemble learning classifiers AdaBoost, bagging and random forest, a relatively new method in machine learning, were trained with 74 features including color histogram features, textures and histogram intersection with immature and mature citrus color model. Overall performances of the three classifiers showed good classification ability for mature citrus (minimum 97% accuracy). Among them, the bagging trees showed the highest accuracy, 91.5, 89.1, 97.4, and 85.2% for green immature, intermediate, mature and diseased citrus, respectively.