Visual Crawling And Trajectory Simulation Of Picking Manipulator Based On Deep Learning

In the field of agricultural machinery picking,robotic arm picking identification and shortest path planning determine picking accuracy.Simple human eye recognition has the factor of repeating mechanical labor,which is likely to cause a decrease in efficiency over a long period of time.However,the use of software or traditional image processing technology can not accurately identify the location of the picking target,or identify the positioning time is too long,can not improve the picking efficiency per unit time.In the complex environment of agricultural fruit forests,it is necessary to have an automatic identification picking operation with high picking precision and low resource energy consumption.Therefore,the research on lightweight high-precision target recognition for picking robotic arms has great market application prospects.This paper mainly studies from three aspects:First,based on deep convolutional neural network target object detection.According to the low power consumption,high speed,light weight,high recognition accuracy and expandable customization requirements in the hardware environment,the lightweight deep convolutional neural network is selected as the kernel,and the multi-platform keras deep learning framework can be selected.The Windows platform or the Raspberry Pi Linux platform uses the standard weight of the YOLO algorithm to perform theoretically fast target classification and recognition for the target in a complex fruit forest environment.The recognition accuracy is increased to 90%,and the time is only 4.94 seconds.Second,robotic arm modeling and trajectory simulation.In this paper,the Epson C4-A601 manipulator is used as the modeling object,and the D-H parameter method is used to establish the linkage coordinate model of the manipulator.The attitude map of the manipulator in the coordinate system is expressed by solving the kinematic equation of the manipulator.Using the homogeneous coordinate transformation formula and the linkage coordinate system,the structure of the multi-link motion trajectory of the manipulator is presented.The picking trajectory path is planned by simulating the kinematic model using the Robotics Toolbox in the MATLAB toolbox.By solving the kinematics problem,the cubic spline interpolation of the fourth polynomial can ensure that the residual value of the trajectory in the simulation process is within the range of(9.39x10-15,1.82x10-14).Arm kinematics correctness and modeling feasibility.Third,visual image processing.The industrial camera captures the captured image for morphological processing,performs watershed algorithm semantic segmentation,canny operator edge detection and Hough edge detection and reprocessing,and obtains the target coordinate parameters for feedback,which helps the robot arm to accurately locate and capture.Finally,using the existing experimental coinditions,the algorithm is debugged.the industrial camera is activated to identify and locate,and the robotic arm grabbing experiment is run.The results show that the robotic arm recognition process takes only 1.5 seconds and the entire grabbing process takes only 13 seconds.The method proposed in this paper can effectively and accurately identify and capture the target object,which greatly saves economic costs and improves production efficiency.