Research On Recognition Method Of Weeds In Cotton Fields Based On Color Feature And Positioning Technology Of Weeds

Weeds in cotton fields have a serious impact on the cotton's yield and quality. Weeds' recognition and their actual characteristic information's determination are premises of effective weeds control in cotton fields.The subject focuses mainly on cottons and their companion weeds,and determines the recognition characteristic parameters and obtains the hidden camera parameters' matrix and extracts the actual characteristic information of weed regions via images' collection and process by Machine Vision technology.The main works are as follow:(1) Recognition method based on the cotton stem's color feature.In the case of cottons' and green weeds' leaves within overlapping but cotton stems' uncovery,the mixed images of cottons and their companion weeds were collected,and plants in the cotton fields were classified into two categories,cotton seedling and green weed. The differences between the dark red feature of cotton seedling stems and the green feature of weeds were sought,and cotton seedling's regions were ascertained by information fusion technology of cotton seedling's location,and then every green weed's regions were identified.The experiment shows that the green weeds are recognized completely,and the recognition rate of the cotton seedling is 74%above under the required conditions.(2) Recognition method based on the overall color feature.In the case of cottons' and green weeds' leaves within overlapping as well as cotton stems' overlapping,Copperleaf Herb in the cotton fields was taken as an example,and the images of individual cotton seedling and individual Copperleaf Herb were collected. It focuses mainly on the overall plants.The gray images and binary images of cotton seedling and Copperleaf Herb were obtained by five methods,the Color-difference's methods(R-G,R-B,G-B),the Super-green's method(2G-R-B),and the Hue's method(H) respectively.The Hue's feature images segmented by Otsu's method could be achieved better results by comparing.Six Characteristic parameters,the standard deviations of R,G and B in the foregrounds(SR,SG,SB) and the standard deviation margin(SB-SR,SB-SG,SR-SG),were analyzed comparatively.In the end, the threshold value for the judgment of the Copperleaf Herb,which was the margin between R's standard deviation and B's standard deviation less than 5,was determined.The experiment shows that the recognition rates of the Cotton and the Copperleaf Herb are 71.4%,92.9%respectively,and the overall recognition rate is 82.1%.The algorithm reduces dependence on the light,and the color features decreases weed identification's error because of within leaves' overlapping and damage.(3) Camera calibration and analysis of weeds' positioning accuracy.The Pinhole Image Formation Model was created based on the low weed positioning accuracy and the experimental conditions' restriction.The images of calibration jig were processed and analyzed,and the hidden camera parameters' matrix and the actual areas per pixel,which is 4.08mm~2,were obtained.The positioning accuracy of targets,orderly weeds and disordered weeds were obtained and they are 0.15cm, 1.92cm and 2.28cm respectively by the hidden camera parameters' matrix.The maximum positioning error,2.3cm,was ascertained after comparative analysis.The obtained positioning accuracy meets the actual required herbicide spraying.The purpose of the subject is to research and develop the automatic weeding equipment for achieving intelligent spraying herbicide.The algorithm of weeds recognition and positioning are optimized and weed recognition rate and computing speed are improved.Those works promote further the research on dynamic process weeds images by Machine Vision.