| Radix Peucedani(Qianhu)is a kind of Chinese herbal medicine with high medicinal value.The lush weeds in the field seriously affect the growth of Radix Peucedani.At present,the planting scale of Radix Peucedani in Ningguo continues to expand,but the level of mechanization and intelligentization of Huhu weeding is relatively low.Aiming at this problem,this article mainly studies the technology of Radix Peucedani weeding based on deep learning.Aiming at the existing problem of weed target detection and location,the method of weed recognition through computer vision technology is used to identify and locate weeds.This paper takes the weeding link in the production process of Ningguo Radix Peucedani as the subject background,and takes the research of weed identification and positioning and weeding as the starting point.The target detection based on deep learning neural network is studied,and the weeding experiment of Delta parallel manipulator motion control is built.platform.This article focuses on computer vision weeding technology,and consults a large number of documents to conduct an in-depth analysis of various weeding principles based on computer vision technology.A data sample set of weeds and Radix Peucedani is established,and a method of weed recognition based on the YOLO V4 convolutional neural network model is proposed.Data enhancement technology increases the complexity of network training,improves detection accuracy,and realizes a vision system for weed recognition and positioning.A method based on the Delta parallel manipulator to realize the mechanical rotary knife weeding on the target is proposed.The kinematics equations and modeling of the Delta parallel manipulator are studied.The hardware control circuit of the Delta manipulator using the STM32F407 chip is designed to complete the production of the overall circuit PCB.According to the mechanical structure principle of the Delta parallel manipulator,the control and communication program of the manipulator was written in C language using keil-MDK software,the selection of electrical equipment was determined,and the Delta manipulator weeding test platform was built.The purpose and requirements of the experiment were determined,and an integrated test platform was established for real-time weed detection,positioning and weeding with industrial cameras,computers,and Delta parallel manipulators,and weed recognition was obtained through deep learning network training.Completed the weed feature extraction based on deep learning convolutional neural network technology,and finally verified the feasibility of the constructed Radix Peucedani weeding test bed based on convolutional neural deep learning. |
