| Tobacco is an important part of Chinese economy.That tobacco maturity detectiontechnology is an important measure to development of Chinese tobacco industry. To solve theproblem in tobacco maturity detection,this thesis employed the image processing technologyin tobacco maturity detection.The main contributies were as follows:(1)The image processing methods for tobacco maturity detection were chosen.This thesisused median filter to implement image denoising. The OTSU thresholding methodwas usedfor image segmentation. Exproment of results showed the image processing can satisfy therquire of this thesis.(2)Single image haze removal using dark channel priorbased on Gaussian filter wasproposed.Using Gaussian filter instead of soft matting to remove blocky effects,smoothtransmission,and recover a haze-free image.Experiment of results showed that the improvedalgorithmcould reduce the amount of computation,achieved20times acceleration than theoriginal algorithm when kept the haze-free image effect.(3)The tobacco color and texture features in different period of tobacco maturity wasdetermined.The tobacco maturity grades could be showed with hue, saturation, brightness,angular second moment, entropy, contrast,correlation and homogeneity.Experimental resultsshowed that the hue, saturation, brightness, angular second moment and correlation wasrising while entropy and contrast was declining when tobacco in tobacco growing process.(4)7-10-3BP neural network classifier structure was Determined.tansig was derterminedas transfer function from the input layer to the hidden layer.purelin was dertermined astransfer function from the hidden layer to the output layer.SCG training algorithm was used totrain7variables of color and texture features.In this way,the maturity model wasestablished.Experimental results showd that the recogniton rate achieved89%.The tobacco maturity deterction model was created by BP neural network with highaccuracy.This thesis provided a new way to detecte tobacco maturity. |
PDF Full Text Request