基于多光谱图像机器视觉的棉田杂草识别研究农业机械化工程专业论文.docxVIP

优秀毕业论文 精品参考文献资料 江苏大学硕士学位论文ABSTRACT 江苏大学硕士学位论文 ABSTRACT The research work on using machine vision system to identify weed is a focus these days as well as a main trend in the future．We regard cotton and cotton field weed as the objen of this study and use multi·spectral images based on machine vision technology to achieve background segmentation,identifying and following process．The main contents of the study are as follows： (1)Background segmentation．Firstly，I establish 2-dimensional histogram by using Near-infrared image and red image and use all segmentation errors as the target to choice segment line．Secondly，I use fisher to reduce dimension and then segment image through the OUSt．Finally，I find that the method of Fisher show better effcet by compraring witll using near-infrared image or the red image alone． (2)Identification features pmposed．Using the ratio of thinning length to leaf area and the ratio of skeleton length to leaf area as two morphology features identify cotton and weed．Using average value of IR、CIR、IR／R these three multi-spectral features identify monocotyrledon and dicotyledonous weed．Using aspect，roundness as well as compactness as three shape features，and standard deviation，smoothness as two texture feature identify weed type． (3)useing method-support vector machine(SⅥ峋based Oil the limited sample and the minimum principle of the structure risk as pattem recognition identify cotton and weeds． Four models with radial basis function(rbO are established in the experiment and then using the grid search method optimize the kernel parameters and the punishment parameter C．the results show that the final correct recognition rate are 98％(cotton)，92％(Setaria viridis)，84％(Eleusine in．ca)，82％(Cephalanoplos segetum)，80％(Portulaca oleracea)．Compared to use shape charactedstics single pattem recognition,total recognition precision has enhanced 12％． The study of iden曲mg cotton field weed will provide technological basis for the further development of precis