Research On Fruit Picking Based On Deep Learning

The project of this paper is a strawberry picking robot project cooperated by Professor Zhao Xiaoguang from The Institute of Automation of Chinese Academy of Sciences and China Agricultural University.The traditional machine vision technology as a detection module of picking robots has some limitations in the detection of targets in the natural environment,which limits the working environment of picking robots.Therefore,a fruit detection algorithm based on deep learning is proposed for picking robots to improve the accuracy of picking.In this paper,the application of deep learning in fruit detection was studied with strawberry fruit planted in greenhouses.The recognition of strawberry fruit and the prediction of fruit rod area were realized by using image processing technology.Deep learning technology can not only overcome the impact of the complex natural environment,but also improve the detection ability of the detection system for the target,which has important production significance for the application of picking robots in multiple scenarios.The main research contents are as follows:? The research results of different fruit detection by domestic and foreign scholars were analyzed,and the application of deep learning technology to strawberry fruit recognition was discussed.This paper further studied the methods of target detection of the "Two Stage" model led by RCNN and the "One Stage" model of YOLO series,and compared the differences in the speed and detection accuracy of the two models.Based on the application scenarios in this paper,YOLOv3 in YOLO series is selected as the basic network.? The strawberry fruit image in the greenhouse was collected and preprocessed.The marked data sets were used to train the YOLOv3 network and the network recognition effect was compared.Combining with the characteristics of strawberry fruit,the improved method of parameter selection and the method of identifying the effective area of the final fruit in the process of network detection was proposed to realize the identification of strawberry fruit in the complex environment.After threshold segmentation,morphological processing and other operations were carried out in the effective region of the fruit to obtain the fruit contour,the region of the fruit pole picked by the picking robot was predicted.Experimental results show that deep learning technology can overcome the influence of natural conditions such as illumination,overlap and occlusion to correctly identify the fruit.The improved method in this paper improves the effect of YOLOv3 model on strawberry fruit recognition.The deep learning technology is used to identify the fruit first and then segment the effective region to predict the region of the fruit rod,which has certain reference value for the fruit with the same growth characteristics.