Study On Rice Quality Detection System Based On Image Processing

The production and consumption of rice has been the most in food crops. Withthe improvement of people’s living standards and the increase of foreign trade, peoplehave paid close attention to the quality of rice. The quality of rice include manyindicators, the appearance of quality directly affect their competitiveness in themarket. Processing quality is the main factor that decides the nutritional value of rice.At present, most of the rice quality detection were completed by artificial,laborious. With the development of information technology and computer application,graphics technology has been used in many disciplines and demonstrated the broadapplication prospects. For a more objective measure of rice processing accuracy,image processing with computer vision technology were used to detect the factors thataffect the quality of rice processing. This article as a part of ji Lin province appliedbasic research project (201205013), mainly for the following research: the broken ricegrain shape detection, machining accuracy detection, transparency test.1The broken rice discrimination model was established based on the grain shapecharacteristics. Rice image acquisition device was used to collect the rice image,using MATLAB for image preprocessing, to extract the grain shape characteristicparameters. Two methods were used for the broken rice shape detection. First, grainshape feature parameters principal component analysis with software SPSS.As such,we can abtain the main components of2features. The accuracy of testing by applyingthe Bayes stepwise discriminant analysis can be as high as94.5%.The accuracy oftesting by applying the PNN neural network is99.4%，higher then the Bayes stepwisediscriminant analysis. Second, the broken rice discrimination model was establishedbased on Fourier transform coefficient method. Edge detection curve of whole rice isrelatively smooth while edge detection curve of broken rice is sharp at inflection point.Get the edge detection curve then Fourier transform, extract10Fourier transformcoefficient, finally get the broken rice discriminant function of bayes discriminantgroup. The discriminant accuracy of whole rice and broken rice were93.3%and98.3%, respectively, the average accuracy is95.8%.2Analyse the relevance between the features of image texture and rice processingprecision. Texture analysis include a variety of methods. Such as statistics, spectrummethod, structure method and so on. This article mainly uses the histogram statistics, gray level difference statistics, gray gradient co-occurrence matrix, run lengthstatistics four methods for texture analysis. After the image preprocessing, thenextract image texture feature of the different machining accuracy of rice based on fourkinds of texture analysis method. After all, to establish texture feature discriminantmodel of rice processing precision, Using stepwise discriminant analysis method andPNN neural network structure, based on data analysis software of SPSS andMATLAB platform. Experimental results show that the rice processing detectionaccuracy of stepwise discriminant analysis with four methods, respectively is:95.625%,84.1%,96.88%,84.1%. The accuracy of testing by applying the PNNneural network with four methods, respectively is:90%、82.5%、90%、82.5%. Sousing histogram statistics or gray gradient co-occurrence matrix texture analysis toextract rice texture characteristic value, stepwise discriminant analysis to inspect therice processing precision is reliable.3Analyse the relevance between the eigenvalues of rice transparency and starchcontent after rice transparency detection based on image processing. Rice imageacquisition device was used to collect the rice image, using MATLAB for imagepreprocessing then extract the rice transparency eigenvalues. Stepwise linearregression was used to construct rice transparency prediction model. Ricetransparency by sensory score as the dependent variable, the image characteristicvalues for transparency as the independent variable. The R2of rice transparencymodel prediction can be as high as0.904.Double wavelength measurement methodwas used to measure rice amylose content, amylopectin content and proportion ofamylose and amylopectin. Analysis the correlation of transparency eigenvalues andstarch content, the results showed that amylose content is related to the characteristicvalue of transparency (m (average grey value), h (color than)) significantly. Amylosecontent prediction model was established with multiple linear regression analysis, riceamylose content as the dependent variable, transparency characteristic value (m, h) asthe independent variable. The simulation effect is not ideal with R2of rice amylosecontent model prediction0.61.4Design the rice quality detection system according to the thought of modernmechanical automation, the electromechanical integration control system was appliedto the rice quality detection. It was achieved that rice of automatic separation,automatic focus and access to rice images fast and efficient. The device containedacquisition device and actuator. Acquisition device was consisted of sealed box, illuminating system, industrial camera, computer, cable and other components. Theactuator contained hoisting devices, horizontal rotary motion device and automaticcontrol system. Also a tray was designed with a lot of oval groove, it can realizeautomatic separation of rice grains, the problem of grain adhesion was solved whenimage analysis. Rice quality detection system was developed based on MATLABplatform, has realized the broken rice shape detection, rice processing precisioninspection, rice transparency detection.