The Key Technologies Of Automatic Picking For Citrus In The Natural Environment

China is an important producer of citrus species in the world,with total citrus production reaching 41.381 million tons in 2018.Picking is the process of citrus production,the largest demand for labor,currently citrus is still mainly harvested by hand.With the rapid development of our countrys urbanization process,agricultural labor costs will rise rapidly,agricultural labor supply will decline rapidly,so the use of automation,intelligent equipment instead of agricultural labor,will be able to solve the labor shortage problem.Citrus picking robots suitable for the natural environment are typical automated and intelligent equipment,which is of great significance in real-time fruit picking,improving the quality of picking and improving the competitiveness of the fruit market.Citrus harvesting operations operate in an unstructured natural environment.There is a complex spatial location relationship between the citrus fruit growing in the natural environment,the fruit and the branches and leaves of the fruit tree,making the fruit,branches and leaves arbitrarily shaded between the existence of randomness,while the uncertainty of light due to changes in time and weather,also makes the characteristics of the citrus fruit change accordingly.The fruit target identification and positioning technology provides the accurate position of the target citrus for the picking robot,which is the basic guarantee for the effectiveness of the picking operation of the picking robot,so in the natural environment,how to effectively identify the citrus fruit and accurately position its spatial position is the key technology to improve the usability of citrus automated picking system.In this paper,the following research was carried out on the identification and localization of citrus fruits in the natural environment.(1)To address the problems of light changes,bright spot shadows and blocking in the natural environment,this paper adopts the design approach of separating citrus recognition and fruit localization,and designs the workflow scheme of the citrus picking point recognition system based on deep learning and feature analysis and the fruit localization system based on visual servo laser directional ranging,respectively.(2)A citrus fruit recognition module based on the YOLOv2 convolutional neural network target detection principle is designed,and a citrus fruit training dataset in natural environment is constructed,and the network weighting parameters of the citrus fruit recognition module are experimentally designed.Because the fruit area identified by the YOLOv2 citrus identification method under partial shading was incomplete,there was a large error in picking point location acquisition.Therefore,a citrus fruit segmentation module based on an improved U-shaped full-convolution network is designed,and a citrus fruit segmentation training dataset is constructed.The network weighting parameters of the citrus fruit segmentation module are experimentally designed to achieve pixel-level segmentation of the identified fruit regions,obtain effective pixels of citrus fruits,and then restore the real citrus fruit contours by Hough transform circle fitting to obtain accurate picking point information.Experiments showed that the accuracy of citrus fruit identification segmentation was 86.5%,and the average distance error rate of picking point location acquisition was about 4.8%.(3)A spatial positioning method for citrus fruits with visual servo laser ranging was designed by combining machine vision principle and sensor perception technology,and the hardware selection scheme,mechanical structure scheme,algorithm architecture and algorithm module implementation of the positioning system were completed,and the overall accuracy and effectiveness of the citrus fruit identification system and positioning system were verified according to experiments.Experimental results show that the positioning system can accurately obtain spatial coordinates of citrus,exclude feasible extraterritorial unpickable citrus,and calculate the optimal picking order to guide the picking system to achieve automatic picking function.The average threedimensional coordinate positioning accuracy of citrus fruits was(±0.8cm,±0.8cm,±1.1cm),the picking success rate was 89.3%,and the picking order optimization algorithm improved the picking efficiency by 34%.In this paper,we construct a method for the identification and positioning of citrus fruits in the natural environment by separating the design of citrus identification and fruit positioning,which has a certain robustness in the presence of light shift,bright spotted shadows and interference from branching and fruit blocking,and has a certain supporting effect on the realization of automatic citrus picking technology.