Research On Siraidia Grosvenrii Features And Spezies Identification Based On Digital Image Processing And Artificial Neural Network

Siraidia Grosvenorii distributed mainly in Guilin,Guangxi province, is peculiar to China as a kind of traditional medicinal plant. Siraidia Grosvenorii now has suffered from mix of variety and degeneration of genetic characterization ,which due to the pursuit of economic benefits. In this paper, species features of Siraidia Grosvenorii are extracted by digital image processing technology. Then the relationship between surface characteristics and internal qualities of Siraidia Grosvenorii is found. Lastly automatic identification of species of Siraidia Grosvenorii is achieved based on artificial neural network . The main completed tasks are as follow:1. Computer vision hardware system that is used to acquire digital image of Siraidia Grosvenorii is designed, which can not only provide stable illumination, but also reduce noise as much as possible. The system can also be applied in other crops.2. The pre-processing program to the images of Siraidia Grosvenorii, such as image transformation, image enhancement, image segmentation and so on is accomplishd under the Microsoft Visual C++. Based on that, the six color features of Siraidia Grosvenorii : Red, Green, Blue,Hue, Saturation, Intensity and eight shape features: Perimeters, Area, Long axis, Short axis, Equivalent diameter, Shape parameter, Elongation, Compactness are extracted.3. 200 pictures about 5 spezies of Siraidia Grosvenorii are pre-processed. Then, Fourteen feature parameters of Siraidia Grosvenorii are extracted from every picture. Through analysing the distributed situation about the same characteristic parameters, we could get the result that different species can be identified through the external features of Siraidia Grosvenorii4.The basic principles about two kinds of artificial neural network—BP network and RBF network are studied. Then the two kinds of network are established in Matlab 6.0. At the same time, the PCA algorithm is used to optimize network input vector in order to reduce network burden. At last, the best network parameters is ascertained after repeated experiments . The average recognition rates of this artificial neural network are: BP:80.91%;RBF:92.68%;PCA_BP:78.15%;PCA_RBF:86.65%。5. In different growth periods, if A represents the red of Siraidia Grosvenorii peel colour, B represents the content of mogrosideⅴin fruit, C represents the hue of Siraidia Grosvenorii peel colour and D represents the content of flavone glycoside , through researching the regularity for change of colour on Siraidia Grosvenorii peel, the conclusion is reached that the regularities of A and B are contrarily; the regularities of C and D are similar . After the experiment , the correlation between color features of Siraidia Grosvenorii peel and the content of mogrosideⅴand flavone glycoside are confirmed.