Research On Target Recognition Of Automatic Disassembly Container Lock Pin Based On Deep Learning

The increasing maturity of industrial automation technology is driving the construction of automated terminals,but the precise identification and disassembly of container locking pins has always been a roadblock on the development path of terminal automation.The traditional way of using the robot arm to disassemble the lock pin is relatively single.The target positioning accuracy is poor and the trajectory planning is not good,which leads to the robot arm disassembly accuracy is low.Therefore,in order to solve the problem of poor accuracy between target recognition and trajectory planning,this thesis studies the target identification and trajectory planning for automatic container disassembly locking pins.The main contents of the thesis are as follows:The main content of this thesis is to study the target recognition and trajectory planning of the automatic disassembly and assembly lock pin of the container.Aiming at the problem of lock pin image recognition and trajectory planning in the task of automatic disassembly and assembly of lock pins,this thesis proposes corresponding solutions.The application environment of the lock pin is complex.The recognition is difficult,and the existing recognition methods mainly rely on large-scale data,which are difficult to apply to working conditions.The trajectory planning research of the robotic arm is to seek the optimal trajectory in the restricted space.First,the thesis introduces the background and significance of this research.The construction trend of automated terminals requires unmanned automation for all types of work.At the same time,the current situation of robot trajectory planning and target positioning is introduced.The thesis also introduces the kinematics analysis of the robotic arm.The thesis mainly describes the basic position and posture parameters of the robotic arm,and analyzes the parameters of kinematics.Then the thesis uses Denavit-Hartenberg parameter method to establish a mathematical model,and express the parameter information of the manipulator in the reference coordinate system through link parameters and joint variables.Finally,the forward and reverse motions of the robotic arm are analyzed.Then,in order to solve the problems of complex working conditions and low recognition accuracy of lock pin image recognition,this thesis proposes an image recognition method based on conditional information deep convolution generative adversarial network.It introduces the generative adversarial network and convolutional neural network in principle,and proposes an improved generative adversarial network based on conditional parameters and information parameters,so that the network model adds guidance parameters to the unsupervised basis to make the accuracy more stable.Finally,it discusses two types of Convolutional Neural Networks(CNN)and Condition Information Deep Convolution Generative Adversarial Network(C-Info-DCGAN)based on parameters such as network structure and loss function.It fully shows that the improved algorithm proposed has a positive effect on the field of image recognition research.Next,for the research on the trajectory planning of the manipulator,the thesis mainly introduces the basic methods of trajectory planning.At the same time,it uses the compound multi-stage polynomial trajectory planning optimization method to strengthen the stability and continuity of the trajectory movement,so that the manipulator runs more smoothly and continuously.The last part is the experiment and simulation.This thesis conducts experiments and simulations on image recognition and trajectory planning.In image recognition research,the collected lock pin data set is used as input data,and the comparative analysis of convolutional network and improved generative adversarial network is used at the same time,which fully illustrates the superiority of improved network for image recognition.Then,through the Matlab simulation software,the joint variables of the manipulator are subjected to the experiment of the compound multi-segment polynomial optimization method to verify its correctness and effectiveness.