Research On Key Technology Of Target Recognition And Positioning Control Of Pleurotus Ostreatus Picking Robot

With the continuous popularization of industrialization and greenhouse cultivation models,edible fungi have become one of the leading industries in agriculture.Among them,pleurotus ostreatus is one of the most common and widely consumed varieties.However,due to changes in light,temperature and humidity in the greenhouse,the growth cycle of pleurotus ostreatus is unstable and the yield is large.At present,the collection of pleurotus ostreatus is basically manual,with low level of automation and intelligence,which is time-consuming and labor-intensive,and the growth environment of pleurotus ostreatus is dark and humid,which will harm people in the long run.Therefore,it has great significance to develop an intelligent pleurotus ostreatus picking robot for picking instead of artificial picking.In this paper,using pleurotus ostreatus planted in greenhouses as the research object to design the pleurotus ostreatus picking robot and its vision system.In addition,focusing on the technology of recognition and positioning of pleurotus ostreatus based on machine vision.Based on this,the pleurotus ostreatus automatic picking control system was designed.The main contents are as follows:(1)First of all,a demand analysis was performed on the planting scene,research object,and automatic picking of pleurotus ostreatus.Confirming the overall goal,main technical indicators and overall plan,then analyzing its working principle.Based on this,the whole structure of the pleurotus ostreatus picking robot and its vision system were designed.(2)Aiming at the target recognition of pleurotus ostreatus in the complex background of the greenhouse,a deep learning detection method based on improved SSD-MobileNet was proposed.First of all,analyzing the characteristics of three forms of deep learning network and three commonly used target detection algorithms:Faster R-CNN,YOLOV3,and SSD;Then building a greenhouse mushroom data set;Based on this,a greenhouse pleurotus ostreatus data set is constructed,and an improved SSD-MobileNet target detection algorithm is studied:first,analyzing the shortcomings of the classic SSD network,combining the characteristics of the deep separable convolution of the MobileNet v1 network,generating the SSD-MobileNet target detection algorithm,optimizing the aspect ratio of the anchor frame according to the shape and size of mature pleurotus ostreatus;Finally,the trained model was used for recognition experiments,and compared with other commonly used models,the results show that the method identified F value is 93.7%,and the detection time of a leaflet is about 0.032s,which meets the recognition requirements of pleurotus ostreatus in greenhouse.(3)Aiming at the three-dimensional spatial positioning of the target of pleurotus ostreatus in greenhouse,the D-H modeling analysis was first performed on the picking robot to obtain the motion relationship of each unit during the movement of the picking robot;Then calculating the 2D center point of the target area of pleurotus ostreatus according to the three-dimensional spatial positioning principle,and using the "8-field"mean method to obtain the 3D coordinates of pleurotus ostreatus with the camera center as the origin in combination with the depth image;Then analyzing the special target traversal method of the camera and performing hand-eye conversion system calibration.The test shows that the positioning success rate is 91.2%,and the positioning error is within 4mm,which meets the positioning requirements of pleurotus ostreatus picking robots.(4)Aiming at the system control and multi-picking unit coordinated control,first,the software and hardware control system of pleurotus ostreatus picking robot was designed,then,focusing on the multi-axis coordinated control of the picking robot,including:target grasping logic planning,and the picking unit’s Motion logic planning,etc.Based on this,the whole machine was tested.The test result shows that the target recognition accuracy of pleurotus ostreatus is 95%,the success rate of picking is 91.4%,and the average time of single target picking is 8.85s,which meets the relevant technical indicators and realizes the greenhouse pleurotus Automated picking,promoting the intelligent production of pleurotus ostreatus.