Design And Target Detection Of Apple Picking Robot

China is the largest country in the world,accounting for more than half of the apple planting area and production.Apple as the king of fruit,it has good taste,rich nutrition characteristics,popular with the public.However,at the present stage,the main picking method is still manual picking.Manual picking is intensive labor,with low picking efficiency and long picking cycle.Therefore,picking robot as a representative of mechanization and intelligent picking will become an important means of production in the future.In this study,three degrees of freedom manipulator with cartesian coordinates is used as the actuator,STM32F103c8t6 as the controller of the actuator and as the lower computer,the computer as the upper computer,the depth camera as the picture collector and depth sensor.The main research contents are as follows:(1)Mechanical arm design research.In order to better adapt to different heights of apples,the height of the picking robot can have a space to move.The three degrees of freedom manipulator arms are all mobility degrees of freedom,respectively on the XYZ axis.The end actuator of the manipulator adopts a claw shape structure,which can wrap the whole apple when picking,and finally carry out the corresponding kinematics equation calculation.(2)lower machine design and programming.The control code of the manipulator is written.The power of the three degrees of freedom of the manipulator is the stepper motor.The starting and stopping of the stepper motor will produce a large impact force on the mechanical arm.In order to reduce the impact force,the starting and stopping of the stepper motor should reduce the speed.The trapezoidal acceleration and deceleration can effectively reduce the impact force.Write an interface,the interface is through serial communication,and can communicate with the computer,to ensure that the computer can and the next machine normal communication,and can give instructions and obtain the status of the picking mechanism.(3)YOLOv4 neural network recognition research.In this paper,the YOLOv4 neural network is adopted and the language used is Python,which can identify apples well.However,the recognition speed of YOLOv4 is low,which cannot meet the demand of real-time picking.Therefore,the neural network is modified by lightweight to achieve a higher recognition speed.The modified network can meet the demand of real-time picking.(4)Research on Apple video recognition technology and 3D positioning.Capture the video and depth frame with the depth camera,and align the video and depth frame.The position of Apple on the XY axis of the depth camera can be calculated by the depth information in the video frame and the compensation and focal length of the offset on the XY axis of the camera.And add position compensation.(5)Robot principle prototype system experiment.This system adopts hierarchical control mode of upper computer and lower computer.Upper computer includes neural network recognition,control algorithm and strategy of manipulator.The upper computer recognizes the apple and obtains the three-dimensional position of the apple in the camera,calculates the position difference between the apple and the robot arm of the picking robot,and sends instructions to the lower computer through the serial port to control the movement of the robot arm.The experimental results show that the average picking time of a single apple in this paper is 12.2 seconds.