Research On Target Recognition And Location Of Apple Picking Robot Based On Improved YOLOv5

In recent years,as the application of robot technology in agricultural picking has become more and more extensive,we have higher requirements for the picking efficiency of picking robots.Apple picking robots can improve the efficiency of apple picking on the basis of reducing the labor intensity of wo rk ers.Fro m t h e a n al y si s o f cu rren t t e ch n o l o gi cal res ea rch p ro g res s,d eep l earn i n g t ech n o l o g y i s in cre as i n g l y ap p li ed i n ap pl e p i ck i n g.Ho wev er,t h e Ap p l e reco g n i t i o n m o d el b a sed on deep learning still has the problems of low recognition accuracy and poor positioning accuracy.These problems have always been a major difficulty in the research of apple picking robot.In view of this situation,In view of this situation,this paper conducts research from the aspects of visual recognition,target positioning and software design,so as to provide a research basis for apple picking robots.The main research contents and conclusions of this paper are as follows:After setting the relevant parameters of the network model,the classical network model SSD,YOLOv4 and YOLOv5 were compared under the same conditions.The results show that the YOLOv5 network model performs well in terms of speed and accuracy.Therefore,YOLOv5 is selected as the basic network of apple identification detection for improvement research.The paper proposes an improved network model based on YOLOv5 and is named YOLOv5-CE.In the feature extraction part of the model,the Conv Ne Xt network with better performance and the attention mechanism are added in the feature extraction process,which increases the m AP of the YOLOv5-CE network model from 92.02% to 93.72%.Calibrate the RGB-D camera Intel SR305 to obtain the parameters required for coordinate conversion.On the basis of completing the camera calibration,an Apple identification and positioning experiment based on Intel SR305 camera was designed.On this basis,the human-computer interaction interface was designed by Py Qt and named Apple Target Detection System.The system can recognize and position apples in three dimensions.Debug and experiment the detection system to verify the effect of the system.The result shows that the designed Apple object detection system is simple and efficient.This paper chooses to improve the convolutional neural network algorithm in deep learning technology.It improves Apple’s recognition accuracy to a certain extent.The visual operation interface of Apple target detection system is also designed,which effectively improves the applicability of the model in practice and provides technical support for apple picking robot.