Research And Application Of Visual Inspection Technology For Texture Of Tabacco Surface

The quality of tobacco leaves has long been measured by gravitational method,which exists large workload,subjective factors,tremendous influence of experience and non-unified grading standard and so on.Appearing inconformable tobacco leaf in the process of acquisition would greatly reduce its products quality and economic benefit.In this paper,the tobacco leaves in the local area of Guizhou Province are used as the research object,studying the color,leaf shape and body shape,surface texture and other characteristics of tobacco samples to extract the best eigenvalue by using machine vision detection technology,based on support vector machine classification technology to achieve automatic grading of tobacco.Specific work of the following aspects:(1)According to the characteristics of tobacco leaves,a tobacco leaf image acquisition device is designed and constructed,and a number of tobacco samples are collected to establish the image sample database of the experiment.(2)Several commonly used pretreatment methods are introduced.There are many types of features extracted in this paper.The preprocessing methods of different feature extraction methods are different.Therefore,we study and determine the pretreatment method which is suitable for the extraction of various characteristics of tobacco leaves.The method is mainly to filter the tobacco leaves to remove the noise.The maximum inter-class method is used to determine and binarize the threshold of the tobacco leaves,including the binarized image except small area.The tobacco leaf image of pretreatment greatly improves the accuracy of feature extraction.(3)The morphological character of leaf is extracted,including leaf length,width,aspect ratio;the body character of leaf extracted,including leaf circumference,area,circularity and breakage rate.Based on RGB color space and HSV color space,the leaf color image is studied.And the color characteristics of R,G,B,H,S and V are extracted.Based on the gray level co-occurrence matrix method,the texture features of tobacco leaf image are extracted,including energy value,entropy mean value and contrast value.Analysis of tobacco leaf surface texture based on Gabor wavelet transform,the texture mean and standard deviation of the image are extracted.And analyzing the data of each type of characteristics,tobacco leaves whose characteristics are helpful for the tobacco leaf grading,are selected as the training parameters of the tobacco leaf grading model.(4)Using the one-to-many classification of support vector machine to construct the tobacco grading model,the model is trained by the characteristic parameters.By comparing the classification model of polynomial kernel function,Gaussian radial basis function and Sigmoid kernel function,the optimal support vector machine model is selected as the tobacco grading model.(5)Based on the GUI platform of MATLAB,the man-machine interface of tobacco leaf quality classification system is developed to realize the human-computer interaction of tobacco leaf in quality grading production line.