Application of image segmentation algorithm based on entropy clustering in apple harvesting robot

For the robot vision system in apple harvesting robot, a new image segmentation method based on entropy clustering is proposed in HSI color space. Firstly, noise was wiped off by using weighted algorithm of median filtering in HSI color space instead of traditional algorithm in RGB model; secondly, Hue and Saturation components were extracted to do entropy clustering with their independence with Intensity, to get an initial segmentation; lastly, the clustering centers were optimized by K-Means clustering, to segment apple object from background correctly and completely. The experiments show that the algorithm can overcome two disadvantages in traditional K-Means algorithm effectively, noise interference and susceptible to the choice of initial cluster centers into local solutions; it can achieve centers automatically, then get an ideal result; the consuming time is short to meet the requirement of real-time ability, the accuracy is high as well.