Inspection Of Longjing Tea Quality By Using Multi-Sensor Information Fusion Based On Hyperspectral Analysis And Image Manipulation

This research uses the multi-sensor fusion technology based on hyperspectral and the imagery manipulation to judge Longjing tea tea quality for the first time. The experiment takes 3 rank of Longjing teas as the experimental objects, which can gain the image characteristic information and the hyperspectral signatures information of Longjing teas. The color color feature and the textural feature of Longjing teas can be obtained by using pictorial information's processing; also some hyperspectral features can be obtained as the absorption depth, the absorption area, the red side position, the red valley position, 6 normalization vegetation indexes and so on. Above all the research fully uses many kinds of sensor's information resources, and obtains the simplified aggregate variable, which constitute the input of pattern recognition. The Longjing tea synthesis judgment model is established based on the theories of genetic algorithm and support vector machines, as the judgment Longjing tea tea the quality standard.Main researches are:1. The image informations of tea samples are obtained. Basic image enhancement techniques are analyzed and applied to the images taken in the quality detection. Kinds of noised are imposed on the image and their corresponding effects are compared. Algorithms of obtaining the color and textural feature are compiled.2. The model primitive spectrum index of reflection curve in the scope of 350~2500nm can be obtained by used Field Spec Pro FR spectroscope. Algorithm related spectrum data pretreatment are compiled, includes data standardization, noise elimination, smooth curve and so on.3. The input parameters of model are choosen. 9 parameters of color and textural, also 8 parameters of spectrum characteristic are choosen as the model input parameters. Support vector mechanism are researched as the algorithm of modeling. The classified experiment is divided three groups to carry on, which are choosen the linear, poly, radial basis function(RBF) and Sigmoid function, four kind of function form's as kernel function to support vector machines to compare the modeling. It's proved that RBF kernel function make the most superior result.4. Genetic algrithm is used to modify parameters of kernel function. The penalty coefficient C and the regularization coefficient Gamma of RBF kernel function are modified.5. IThe theories of the genetic algorithm and the support vector mechanism are researched. Those theories are combined in the application of the determinations of Longjing teas and modeling. It's proposed the genetic support vector machines modeling system. The classified experiment is divided three groups to carry on, which are choosen the linear, poly, radial basis function(RBF) and Sigmoid function, four kind of function form's as kernel function to support vector machines to compare the modeling. Genetic algrithm is used to modify the penalty coefficient C and the regularization coefficient Gamma. Experimental results showed that it was feasible to discriminate Longjing tea's comprehensive quality. The correct rate of testing samples was 89%, and the cross validation accuracy was 100%.