<i>Research On Recognition Methods of Pomelo Fruit Hanging On Trees Base On Machine Vision</i>

摘要

<b>Abstract.</b> <sc>.</sc> Recognizing the pomelo fruit on tree accurately is the presupposition for harvest robots. In this work, 3 recognition methods, including chromatic aberration algorithm, K-means algorithm and Yolov3 algorithm, were applied into recognize pomelo fruit from the images of whole tree taken at three distance, including further than 2.0m, between 1.0 and 2.0m, less than 1.0m. Recognizing model was established by labeling and training with 400 images of QuanShui pomelo hanging fruit trees. Then the model was verificated by distinguishing the fruit from complex background in images taken under different light conditions and at different shooting distances. Depending on the performance in the verification set, Yolov3 algorithm obtained the best results with 94.22%, 91.48% and 80.92% as the accuracy rate of less than 1.0m, between 1.0 and 2.0m and further than 2.0m distances respectively, while chromatic aberration algorithm recognized 58.67%, 53.44% and 39.92% and K-means algorithm distinguished 73.20%, 66.46% and 43.97% respectively. The total accuracy rates achieved by Yolov3 algorithm, chromatic aberration algorithm and K-means algorithm were 88.87%, 50.68% and 61.20%. These results shown that the machine vision could be used to detect the pomelo fruit on tree and Yolov3 algorithm as a deep learning method could improve the accuracy rate obviously.