Research On The Influence Of Different Wave Band Light Sources And Light Intensity On The Classification Of Tobacco Leaves

Tobacco grade quality is related to cigarette aroma,for the current tobacco leaf purchase stations,the tobacco leaves are graded under a single white light source and light intensity environment,in this paper,the classification of tobacco leaves collected under different light sources and different light intensities is studied,in order to select the best light source and light intensity for the classification of tobacco leaves.Main tasks as follows:(1)The tobacco image acquisition equipment was studied.The device can collect tobacco leaves under four light sources of white,red,blue and green,control the light intensity of each light source by adjusting the light intensity controller,the entire collection process is carried out in a closed and opaque environment,as far as possible,ensure that the collected tobacco image is consistent with the sample.(2)Tobacco image preprocessing.Adopt the adaptive correction algorithm of uneven illumination image based on two-dimensional gamma function to realize the illumination equalization;grayscale the equalized image,select 3×3 template median filter for noise removal through experiments?binaryization using maximum inter-class variance(Otsu)?then perform small area removal and hole filling to ensure the integrity of tobacco leaves?finally,the background is removed to obtain a color segmented image containing only tobacco leaves.(3)Tobacco leaf characteristic parameter extraction.Comparative analysis of RGB color space model and HIS color space model,extract the 18 color feature parameters of the consistency U,entropy e,first-order moment ?_i,second-order moment ?_i,third-order moment s_i,and smoothness R of the three channel components under the HIS color space model;the relative energy(e_h~2,e_v~2,e_d~2,e_h~3,e_v~3,e_d~3)of the detail sub-graphs of the horizontal and vertical diagonals of the second and third layers of wavelet is extracted as the texture feature parameter.(4)Sensitivity analysis of tobacco leaf images under different light sources and different light intensities.Applying the improved image information entropy theory and structural similarity theory to the tobacco leaf images under different wavelength light sources and different light intensities,select the best band light source and light intensity.(5)Verify that the model is established.Based on the PCA-BP neural network,a verification model is established to test the effectiveness of the sensitivity analysis method of tobacco leaf images under different light sources and different light intensities.From the classification accuracy and error analysis charts of the middle tobacco leaves under different light sources and light intensities It is effective to apply the improved image entropy and structural similarity theory to tobacco leaf image sensitivity analysis methods with different light sources and different light intensities.The research results of the paper show that different band light sources and light intensity have a significant impact on the feature extraction and recognition process of different grades of tobacco leaves.Choosing appropriate band light sources and light intensity can effectively improve the classification accuracy and effect.Finally,it was concluded that under the white light source,the tobacco leaves collected with a light intensity of 90 lux were the most favorable for classification,and the recognition accuracy and efficiency were higher than those of other band light sources and light intensity.