Research On Target Recognition And Location Of Retractable Picking Manipulator

Gannan navel orange is one of the famous specialty fruits in Jiangxi.It contains all kinds of nutrients necessary for the human body,which can help the body to form cancer suppressing substances,and soften and promote blood circulation,and which enables to protect blood vessels and reduce cholesterol and blood lipids.As a result,it is deeply loved by people.However,Ganzhou terrain is mostly mountainous and hilly.Navel orange harvesting is mainly manual,which takes a long time and has poor timeliness.Considering of the short period of navel orange harvesting and the high labor intensity and input,it has become an urgent problem to develop the automatic and intelligent picking of Gannan navel oranges.In this paper,the Gannan orange picking under the natural environment was regarded as the research object,combining image processing,deep learning,binocular vision,mechanical system design and other technologies,a retractable Gannan navel orange picking manipulator was developed.The main research contents are as follows:(1)Study on the image acquisition and image preprocessing methods of Gannan navel orange.The head up shooting method for image acquisition could eliminate the interference of factors such as the sky and land.Comparing the color effects of navel orange images under the three color space of RGB,HSV,and Lab,the color images in the RGB color space were significantly different from the background.Therefore,navel orange color image samples in RGB color space were used in this study.In order to highlight the color and texture features of the original images,median filtering,wavelet analysis and guided filtering were used to remove the noise interference from the original image.The results showed that the guided filtering had the best denoising effect,which could better remove the extra noise information in the image,highlight the navel orange color and texture features,and improve the contrast with the background.(2)Study on target recognition methods of Gannan navel orange under natural environment.In this paper,deep learning methods was used to identify navel orange fruits,and the two-stage target detection algorithm Faster R-CNN and the single-stage target detection algorithm YOLO v3 were chosen to conduct recognition model training for 1100 navel orange image data sets.The loss values during the model training process were statistically analyzed,and 20 navel orange images were used to verify the model performance.The results showed that the Faster R-CNN model had fewer training times,and lower loss value when it was stable,and better performance on identifying orange fruits with overlapping and shaded by branches and leaves.The precision of Faster R-CNN model was 86.55%,and the recall was 98.56%,and the F1 score was 92.17%,exceeding that of YOLO v3.(3)Design of retractable picking manipulator for Gannan navel orange and target recognition and location of manipulator.In this paper,the retractable picking manipulator for Gannan navel orange was designed,which included three parts:mechanical structure,vision system and control system.It mainly completed Gannan navel orange image samples collected,pretreatment,target recognition,location and picking.And the binocular vision system was used to realize binocular calibration,binocular correction,stereo matching,and three-dimensional spatial positioning of the navel orange,and further the 3d spatial coordinates of navel orange were reconstructed.The control system carried out the inverses kinematics of the manipulator on the three-dimensional reconstructed coordinates,and sent the results of the inverse solution to the lower computer through the serial port,and controlled the picking actuator to reach the location of the navel orange to complete the picking work.In the laboratory environment,the accuracy of the identification and location of navel oranges picked by the manipulator were tested.The results showed that the recognition rate of the manipulator for the Gannan navel orange reached 93.63%,the success rate reached 83.63%,and picking time efficiency is 8s each.The retractable picking manipulator for Gannan navel orange developed in this paper can realize the automatic picking of navel orange,and provide a basis for the picking machinery for navel orange and other crop.