Research On The Seed Information Of Rapeseed Based On Digital Image Processing

The thesis makes a study on a quantified way for rapeseed's shape, color andweight information based on computer vision and image processing technology. Thestudy involves following works:(1) For the standardization of digital images photographed, the correspondingshooting specification is drawn up. We choose a system calibration method with VCDdisc as a point of reference, and image preprocessing methods with gray and medianfiltering.(2) By comparing 10 kinds of image segmentation technology in three groups, wechoose OTSU global thresholding and two apex thresholding as the main imagesegmentation means.(3) The round degree and radio of short axis to length axis are taken as theclassification parameters. Three varieties of rapeseed are classified and the digitalquantity parameters are extracted by the fuzzy C-means clustering method, based onwhich the quantity is given to regulate the rapeseed shapes. Furthermore, the rapeseedgrade is discussed. The reliable foundation for rapeseed classification andidentification is provided.(4) The coefficient variation of radius is used to estimate the plumpness ofrapeseed, and establish the digital model of plumpness in single rapeseed. What'splumpness is greater than or equal to 0.6 is full. The plumpness ratio is the bestindicator of full rapeseed.(5) The equivalent diameter is used to estimate the weight of rapeseed, andestablish the special and common model. In the overall, the correlation coefficient ofpredicted and measured values was over 0.99. The accuracy is more than 95%.(6) The nine color HSV model, major color means and yellow rapeseed methodare used to identify and quantify rapeseed's color. It can realize on a single seed coloranalysis for breeder's reference. (7) The equivalent diameter and plumpness ratio are used to identify two classrapeseed, B. compestris and Brassica napus L., accuracy is 100%. With equivalentdiameter, plumpness ratio, major color ratio and the mean degree of yellow rapeseed,the second classification is used to distinguish between six varieties of rapeseed; therecognition rate can reach 95%.(8) By comparing the performance of stepwise discriminated analysis, BP neuralnetwork and SVM, we find the result of BP neural network is optimal.(9) To facilitate computer analysis of above issues, the digital image processingsystem for the field of agriculture was developed by use of MATLAB. It can providethe morphology, color and weight information of rapeseed through image analysis. Itcan also be used to obtain other information, such as leaf area, leaf color. The softwarecopyright has been declared from the State Copyright Bureau. And the softwareapplication performance is significant in rice (see another master's thesis).