Identification of fruit and branch in natural scenes for citrus harvesting robot using machine vision and support vector machine

Abstract: With the decrease of agricultural labor and the increase of production cost, the researches on citrus harvesting robot (CHR) have received more and more attention in recent years. For the success of robotic harvesting and the safety of robot, the identification of mature citrus fruit and obstacle is the priority of robotic harvesting. In this work, a machine vision system, which consisted of a color CCD camera and a computer, was developed to achieve these tasks. Images of citrus trees were captured under sunny and cloudy conditions. Due to varying degrees of lightness and position randomness of fruits and branches, red, green, and blue values of objects in these images are changed dramatically. The traditional threshold segmentation is not efficient to solve these problems. Multi-class support vector machine (SVM), which succeeds by morphological operation, was used to simultaneously segment the fruits and branches in this study. The recognition rate of citrus fruit was 92.4%, and the branch of which diameter was more than 5 pixels, could be recognized. The results showed that the algorithm could be used to detect the fruits and branches for CHR.