Research On Key Technologies Of Precise Spraying Of Fruit Tree Spraying Robots Based On RGB-D Camera

With the rapid development of agricultural modernization,the fruit tree spraying robot gradually replaces the inefficient traditional artificial spraying method.However,the existing fruit tree spraying robots are mainly used by constant and undifferentiated spraying methods,which has the problem of excessive pesticide spraying,which leads to its application and development are greatly limited.And pesticide precision spraying is an effective way to solve the above problems,so this paper studies the key technology of precision spraying,focusing on visual sensor and pulse width modulation(Plus Width Modulation,PWM)variable spraying control technology,deep camera for detection equipment,design based on the variable spraying system of the camera,in order to improve the utilization of pesticide fruit tree spraying robot,reduce the dosage.The main research contents and results are as follows:(1)In this paper,the citrus garden on the hilly landform is taken as the practical application scenario to design the variable drug spraying system scheme.Firstly,through the field investigation of citrus orchards,the basic data of the height and spacing of fruit trees are collected to provide data support for the design of robot size and hardware selection;secondly,the design scheme of variable spraying system based on depth camera is proposed according to the functional requirements of the variable spraying system;finally,the spray Angle and layout spacing are determined through the nozzle determination and layout test,which meet the operation requirements of effective spraying.(2)In view of the problem that it is difficult to accurately detect and segment the canopy of a single fruit tree in a complex environment,this paper takes citrus trees in different growth periods as the test object,and proposes a fruit tree canopy image instance segmentation model based on improved Mask R-CNN.First,the D435 i depth camera was used to collect canopy color images and RGB-D images and serve as the data set required for model training;Second,to further improve the model performance,MSEU R-CNN model was proposed by improving the network structure of Mask R-CNN;Finally,MSEU R-CNN and other models were compared to verify that it is advanced in the task of citrus canopy instance segmentation.(3)In order to calculate the real volume of the canopy according to the pixel value of the segmentation mask,this paper based on the canopy mask results generated by MSEU R-CNN and uses the principle of depth camera ranging to build a visual detection crown volume model,and then proposes a citrus canopy volume detection method based on depth camera and segmentation mask.The manual measurement method and the fitting measurement method are used as the comparison method of the measurement accuracy of the crown volume measurement of the citrus canopy.(4)In order to verify the effectiveness and practicability of the variable fruit tree spraying robot spraying in this paper,the spraying performance test was carried out.Firstly,by building the spray test platform,the static PWM variable spray control test is conducted,so as to determine the relationship between the PWM duty cycle,spray pressure and nozzle flow;in the second,based on the design of the variable spray system,the fruit tree spray robot is set up;finally,using the robot,the robot and the target spray are in full.