Research On Rapid Detection System Of Soybean And Corn Based On Machine Vision Technology

This experiment took 4 kinds of soybean(U.S. Soybean, Argentina Soybean,Brazil Soybean, Heilongjiang Soybean) and 2 kinds of corn(Zhengdan958, Hel iongjiang corn) as research materials, and under a certain experimental conditio ns(25℃,black box,a single light resource) to simulate the process of rapid deter mination on soybean moisture content by machine vision technology. During th e experiment, using power-shot camera of fixed and consistent parameter to sh ot the entire soybean and corn samples. Extracted 24 digital image features thr ough programming by MATLAB. Established a discriminant model through the correlation analysis of the 24 features and moisture content by using SPSS an d MATLAB etc. Finally, through the MATLAB GUI(graphical user interface)design and develop a set of online moisture content of soybean moisture meas urement software. The results showed that:(1) The moisture content of soybean samples increases follow the increase of RGB values, with the extension of time, the color, saturation of soybean a ll decreased. The detection of the same origin and batch soybean moisture cont ent is reliable by using RGB values as the factors to influence moisture conten t. With the change of soybean moisture content, the measurement of the accura cy of the software will be decreased to some extent. According to the correlati on analysis between soybean moisture content and RGB values established mult iple regression equation,the result showed that :Heilongjiang soybean RGLA m odel’s regression equation was Y=-15.353+2.738XR-0.325XG+5.028XL-1.238 XA,R2=0.987;Heilongjiang soybean HSLA model’s regression equation was Y=-105.050+0.748XH-16.517XS+29.239XL-4.360 XA, R2=0.994。After recombination betwe en the color characteristic value model of RGB, HSV of Heilongjiang soybean and the contour characteristic value, the fitting degree had enhanced significan tly. The prediction accuracy of the RGLA model had improved 9.5% compared with the RGB model; The prediction accuracy of the HSLA model had impro ved 2.5% compared with the HSV model. The RGLA model’s regression equation of America soybean was Y=-18.625+0.113XR+0.002XG+4.54XL-1.068 XA, R2=0.985; The VLA model’s regression equation of America soybean was Y=-13.303+16.584XV+3.987XL-0.877 XA, R2=0.987, the predicting precision in RGLA m odel and VLA model had increased by 18.1% and 1%.(2) 24 corn digital image features were captured through programming by MATLAB. The results showed that: each of the 24 corn image features could not be the unique factors that predict the corn moisture content. The single cor n image characteristic value and the corn moisture content had no significant c orrelation. According to the size of correlation, recombinated the characteristic value of two producing areas corn, the results showed that, the RGB model’s r egression equation of Henan corn was Y=11.642+0.035XR+0.019XG+0.057 XB, R2=0.974; the HSV model’s regression equation of Henan corn was Y=14.188+8.862XH-0.094XS+3.657 XV, R2=0.977; the R2 of contour characteristic value model LAR was 0.650; the RGLA model’s regression equation of Henan corn was Y=22.603+0.046XR+0.066XG-3.007XL+0.516 XA, R2=0.980; the HVLA model’s regression equation of Henan corn was Y=53.274+10.569XH+1.633XV-9.508 X L-1.547 XA, R2=0.990. The RGB model’s regression equation of Heilongjiang c orn was Y=10.301-0.047XR+0.130XG+0.046 XB, R2=0.919; the HSV model’s re gression equation of Heilongjiang corn was Y=8.714+23.246XH+6.500XS-13.445 XV, R2=0.967; the R2 of contour characteristic value model LAR矩was 0.588;the GLA model’s regression equation of Heilongjiang corn was Y=-23.577+0.108XG+17.258XL-7.786 XA, R2=0.919; the HSLA model’s regression equation of Heilongjiang corn was Y=49.440+15.733XH+7.689XS-61.855XL+38.496 XA, R2=0.968.(3) The experiment had analysis systematically all the images of corn and soybean through MATLAB programming. Established a system named Food Moisture Content Rapid Detection System based on MATLAB GUI(graphical user interface). This system had following functions: the type of image transfor m, image feature extraction, grain moisture content calculation and curve drawi ng. The system could record the change of grain moisture content, which fromdifferent producing areas and type of grain. Through link the database, the syst em could record rapidly, update conveniently on different types data of grain.Through the verification of the software, the result showed that: the prediction accuracy on soybean and corn could run up to 95%.