Research Of Southern Sinkiang Jujube Color Classification Method Based On Machine Vision

This thesis junzao jujube and ash in southern Xinjiang as samples, combined with national standardGB/t5835-2009-date appearance classification method of detection using machine vision inspectiontechnology and digital image processing technology for image acquisition, image pre-processing andanalysis of extracting jujube colors, textures, and defect characteristics based on BP neural network method.Exploration carried out a study on the quantitative classification of external quality of jujube in Xinjiang.Main using MATLAB prepared of image pretreatment algorithm on collection of dates image forprocessing, extraction dates of color features parameter for BP neural network entered layer, achieved datesquality grading; using RGB model and HSI model, respectively extraction dates image in the extraction redbody are value (R), and green body are value (G) and blue body are value (B) and they of are variance,and, and total6species color features variable; again will image from RGB to HIS color space conversion,then from HSI color space in the, Extraction chroma are value (H), and brightness are value (I) andsaturated degrees are value (S) and they of respective of are variance, and, and total6a color featuresvariable, total12a color features variable as BP neural network of entered layer, achieved dates colorgrading; for texture, used statistics of method, are value, and standard poor, and smooth degrees, and threeorder moments, and consistency and entropy histogram statistics features as6a indicators, achieved datestexture detection; dates defects detection problem, through gets dates pictures, on original image forpretreatment, Then proceed to image segmentation, feature extraction and image recognition, to obtain thenumber of pixel defects, then get date part of the pixel, and their ratio is the basis of the trial judge andjujube, an implementation flaw removed. Empirical research results are as follows:(1) The date defects detection, disease, pests and in accordance with national standard GB/t5835-2009pulp head and head must not exceed5%. Test surface under a judgment for good dates, judged to be flawedin Burgundy. According to the defect area and surface area ratio to determine the surface integrity of jujubedates accurate up to91.33%, determine the accuracy of defects on the surface of red date for89.33%.(2) Date texture classification, taking300picture dates as set60as a test set. Will extract the meanvalue, standard deviation, consistency and smoothness, third-order moments,6texture characteristic ofentropy as the input nodes, that is, InPoint=6. Identification of jujube is divided into3levels, the outputnode3(OutPoint=3), the middle-tier nodes for MidPoint=(InPoint+OutPoint)/2=4.5to calculateMidPoint=4the same as MidPoint=5grade of accuracy, we take MidPoint=4, learning rates, maximumnumber of trained5,000times, minimum mean square error of0.01. Found: primary date recognition rate is97.8%and secondary date recognition rate is80%, three-stage date recognition rate is95.6%.(3) The date color grading, and HSI color model RGB color model when the classification wasadopted, the jujube is divided into special effects, first grade, second grade, third grade and substandard,premium jujube recognition rate is96%, level, date recognition rate is80%and secondary date recognitionrate is84%, three-stage date recognition rate is94%, such as date recognition rate is98%.