| Cotton weeding robots need to be able to effectively distinguish between cotton and weeds in order to remove weeds in cotton fields and increase the yield and economic benefits of cotton.Since the leaves of a cotton plant are very similar to those of weeds,it is common for cotton to be mistaken for weeds.This paper is a study on how this problem can be solved.In this study,both cotton leaves and all sorts of weed samples were collected from cotton fields.Digital image processing is used to extract the sample blade geometry features,Hu invariant moments and symbiotic torque characteristics.(1)Collected five plants of cotton,black bean,ash,cocklebur,and morning glory,and conducted noise reduction,graying,and binarization,then morphologically manipulated the binary image,and finally used edge tracking.(2)Based on the preprocessed image,the special features,Hu moments,and texture features are extracted.Among them,on the basis of binary images,the four leaf shape features of cotton and weed leaf width-length ratio,elongation,circularity,and density were extracted,and seven HU moments were invariable;on the basis of grayscale images,extraction was performed.Five texture features based on energy,entropy,contrast,correlation,and deficit moments of the gray level co-occurrence matrix.(3)Select the k-neighboring,naive Bayes,BP neural network and support vector machine four classifiers to compare and classify the images of Chinese herbal plant leaves,and choose a classification method with higher recognition rate.The experimental results show that the classification accuracy of BP neural network is the lowest(87.94%),the support vector machine is the highest(98.81%),and the remaining two classifiers can reach more than 92%.Therefore,the experiment proves that the use of digital image processing technology to identify cotton and weeds is feasible;and the classification accuracy and the consumption time,support vector machine is obviously better than the other three classification algorithms. |
PDF Full Text Request