Abstract:In the automatic fruit picking system, it is one of the most important aspects to recognize apples, especially green apples. Quick and accurate identification directly affects realtime operability and reliability of picking robot. In order to realize recognition of green apple in natural illumination condition, images of apple trees in natural growth period were taken, and random forest algorithm was used to classify and identify green apples. To solve complexity and fuzziness of green apples and fruit trees and complex background’s color and texture features, especially similarity of green apples and leaves on many characteristics, the Otsu threshold segmentation method was applied to remove the background noise and tree trunk and branches in images in RGB space so that images contained only green apples and leaves were obtained. After filtering processing on images, the grey level information and texture features of apples and leaves were extracted respectively, and they were used to train and build the green apple identification model based on random forest algorithm. Then green apple prediction experiments were carried out for the sample images by using template pixel scanning, and the predicting accuracy reached 90%. Finally, ten green apple tree images were chosen to execute green apple recognition by using the model, and with Hough transform method to mark the identified apples. It illustrated that the green apple recognition rate reached 88%. The results showed that the method had a good robustness, stability and accuracy, and it could be used to recognize green fruits under natural illumination conditions.