Real-time Weed Identification Technique For Precise Spraying Intelligent Weeding Device Based On Machine Vision

Weeding is an important and laborious task in field management;therefore enhancing weeding automation has a far-reaching significance.Based on the actual application of weeding robot,this research carries out a study on weed real-time identification technique in the cotton field.The research mainly concludes several aspects as follows:1) The design and construct of machine vision system:design,manufacture and assemble the installation frame,shock absorption mechanism,artificial illumination device of machine vision system;choose and construct camera,lens,computer system, battery and lamp.2) The study of background segmentation method:by programming realize the three effective and common gray processing methods(Excessive Green,Normalized Excessive Green and Cr color differential) and four binary processing methods(fixed threshold method,Ostu method,optimal threshold method,iterative threshold method); obtain five background segmentation methods by comparing and analyzing treatment effects,then carry out the contrastive research as well as analyze its adaptability and time-cost,and choose CLD method of good adaptability and little time-cost as the background segmentation method in the following on-line work experiment;contrast the experiment results of indoor static and outdoor dynamic experiments,carry out the analysis of noise types and noise source,choose suitable noise eliminating algorithm to eliminate noise for each background segmentation method,and improve the image quality.3) Weed identification study:extract the plant growing information,identity cotton and weed by using growing information,extract foreground F(plant) feature(ratio of subject skeleton length and subject area) by using skeleton algorithm,identify cotton and weed by using F feature;carry out a contrastive analysis of the identification effect and time-cost about the two identification methods,choose the identification method based on plant growing information as identification method in the following on-line work experiment method.4) Spraying information research and on-line experiment:carry out a study on the spraying strategy and spraying information obtaining according to the spraying system; denote the weed location and intensity information in terms of a two-dimensional array as well as carry out an analysis about the gaining effect of spraying information;install the machine vision system on intelligence weeding robot to undergo on-line work,study and analyze the spraying effect.5) Develop machine vision software:develop the machine vision software in off-line experiment and on-line work,the software has the image processing function commonly used in weed identification,has good man-machine interface and manipulability.This project develops intelligence weeding platform based on which carries out a study of weed real-time identification technique.It enhances the identification precision and algorithm real-time by studying background segmentation method,weed identification method and spraying information.It also applies this real-time technique to intelligence weed robot for on-line work experiment,which provides a foundation for the deep study and improvement of the identification technique based on practicability,and which makes the decisive promotion function for intelligence weeding robot product forming based on machine vision.