Research On Recognition And Shearing Method Of Grape Bunches Fruit Stem Based On Deep Learning

Grapes are rich in nutrition and unique taste.In recent years,the domestic planting area has been increasing year by year,and it has become the pillar industry and the main wealth source of agriculture in Xinjiang and other places.Due to the short grape harvest period,the obvious shortage of labor in the picking season,and the increasing labor costs year by year,the development of the grape growing industry has been affected.Therefore,research on the technology of grape picking robots is of great significance for reducing the labor costs of farmers and alleviating the seasonal labor shortage,promoting the development of the grape growing industry and improving the efficiency of grape harvesting.At present,the research on the related technology of grape picking robot is in the stage of grape identification and positioning,and there is no mature technology applied to grape picking.This paper focuses on improving the recognition rate of fruit stalks and the research of fruit stalk cutting methods,taking red grape grapes as the research object.The main research contents are as follows:1.Due to the complex growth background of stalk grapes,the premise of stalk recognition is to accurately segment the grapes from the background.After studying the image segmentation method based on deep learning,and comparing with the traditional image segmentation method,the DeepLabV3+ model is finally adopted.The DeepLabV3+ model has a recognition rate of 91% for grapes,78% for fruit stalks,22%for grape category segmentation,and 66% for fruit stalks.2.Aiming at the problem of the low recognition and segmentation rate of grapes by the DeepLabV3+ model,based on the color characteristics of the grape image and the principle of color synthesis,the channel parameters of the RGB color model and the circle parameters are innovatively connected,and the RXY model is proposed.After the RGB image is processed by the RXY model,the DeepLabV3+ model is used to segment the grape image.The result is better than the direct segmentation of the RGB image.The grape recognition rate is increased to 95%,and the grape category segmentation error rate is reduced to 16%.3.In view of the problem that the DeepLabV3+ model has a low stalk segmentation rate,which results in low stalk shearing success,this paper proposes a stalk shearing method that recognizes and cuts the existing area of the stalk,which is compared with finding the stalk picking point.Cutting method,the difficulty is greatly reduced.Using the RXY model combined with the segmentation results of DeepLabV3+,the area where the stalk exists is calculated,and the recognition rate of the stalk is again increased to87%.4.Aiming at the problem that the grape branches around the stalk affect the shearing of the stalk,this paper proposes a plan to grasp first,then move and cut last.According to this scheme and combining the characteristics of different shapes of grape bunches,a two-finger pneumatic manipulator with a rotating knife holder is designed.5.For the research of fruit stalk identification and fruit stalk cutting methods,a laboratory picking platform and a field experiment platform were built.A laboratory picking platform was built in the laboratory to study the fruit stalk shearing method,and a simulated picking experiment was carried out.The experimental results showed that the picking rate of grapes in the vertical state was 78%,and the picking rate of grapes in the inclined state was 72%.At the same time,a field experiment platform was built to conduct field tests to verify the effectiveness of the identification and positioning methods.The experimental results showed that the missed grape detection rate was 23%,and the grape positioning error was between 15 and 40 mm.