Design Of Control Software Of Robot Arm Based On Multi-mode Image

Ultrasound images have the characteristics of short inspection time,no harm to the human body,and low inspection costs in medical testing and disease diagnosis,and they are becoming one of the most important diagnosis and treatment methods.In the medical system,the robotic arm is also showing an effect that exceeds that of humans.The use of the robotic arm for ultrasonic detection has high positioning accuracy and can avoid direct contact with the human body.In the face of the global spread of the new crown epidemic,the method of using robotic arms to automatically scan the human body and collect ultrasound images is receiving more attention.Based on the convolutional neural network and reinforcement learning method,this paper designs a software with the core function of realizing autonomous abdominal scanning by a robotic arm.This software can automatically scan the human abdomen and locate the human liver,and complete liver ultrasound images.collection.The software proposes a complete system based on the robotic arm.Through the software,all the tasks that the robotic arm can accomplish include designing paths,replacing path planning algorithms,performing mobile tests,and linking the robotic arm and the camera with visual guidance.The main research contents of this topic are:First introduced the software/hardware platform for software development,analyzed the functional requirements and non-functional requirements of the software,divided the software modules and software levels,introduced the related functions of the software,and proposed two types of software used in this topic The optimization of software is network communication optimization and process interaction optimization.Then designed an environment in which reinforcement learning is deployed in the robotic arm.The environment involves key information such as the setting of the action cycle and command transmission delay.At the same time,parallel computing is designed to accelerate learning.Train the two A-C reinforcement learning methods on the robotic arm and smooth the predicted path,and finally compare and analyze the effect of the traditional path planning algorithm.Last on the basis of the YOLO v3 detection network,the network is pruned according to the small scale and nonlinear characteristics of the use scene of this subject,and the dichotomous cross entropy is introduced to improve the loss function,and training and testing are carried out on the data set of this subject,Analyze the trained model according to the evaluation index.