Detection Of The Maize Qualities Based On Machine Vision Technology

Machine vision technology is used to simulate visual function to achievenondestructive testing, through acquiring and processing images of object to be measuredto get properties of the target. Machine vision technology applied in cereal’snon-destructive testing have the significant characteristics of non-subjective, high-speed,and high-precision of detection, so the theoretical research and application of thetechnology have obtained extensive attention. Especially, with the improvement of thequality of people’s life, whether the corn used for edibility, industrial processing or asseeds for agricultural reproduction, the corn quality are highly demanded more and more.In order to ensure the quality of the corn, it is necessary to adopt effectivelynon-destructive testing techniques.Based on theoretical research and experimental analysis, this article presented athorough study of the automatic detection method of the corn quality and the system ofreal-time analysis of the maize was established. According to test the characteristics ofsingle kernel corn, an approach for inspection of touching corn kernels was developed torealize image segmentation, information extraction and quality evaluation. The maincontents are as follows:(1) Relying on the image acquisition system, the tile corn imageswere collected, so that the information of a single kernel was obtained. After pretreatedoperation of image enhancement, the images were more suitable for calculation.Subsequently, two different methods of image processing including active contour modeland watershed were comparatively studied. Computation speed and segmentation resultsshowed that the watershed algorithm was more suitable for actual application. Then, inorder to achieve successive segmentation and improve the efficiency of imagesegmentation, the alterable threshold method was used to improve the watershed algorithm.At the same time, the improved algorithm preserved the shape of a single grain image.(2) Single grain region was tracked and marked,37feature parameters including shape, colorand texture were extracted to describe the appearance. There was a great connectionbetween the features during pattern recognition which would result in data redundancy. Inorder to avoid this situation, this article adopted genetic algorithms and principalcomponent analysis to deal with the feature data for selecting valid classification featuresas input parameters for pattern recognition algorithm and then compared the advantagesand disadvantages of the two methods. The results showed that genetic algorithms had alarge scope of application and could acquire the importance of individual feature parameter;however, principal component analysis required all data compression and appearedunintuitive.(3) This paper discussed two pattern recognition methods based on SVM modeland BP neural network model and compared the efficiency and precision of the twomethods by actual experiment. The experiment results illustrated that BP neural networkwas more suitable for the identification of defective particles. When optimization algorithmof features and neural network combined, the recognition network had a better convergenceand high identifying accuracy was found. Meanwhile, the thousand-seed weight wascalculated according to the function between corn weight and detected area, the values ofcalculation and actual measurement were compared. The experimental results demonstratedthis method could be used to predict thousand-seed weight for evaluating corn quality.Eventually, this article realized image processing function described above based onMatlab software and a decision system based on Labview was built to develop a good userinterface and the communication of these softwares was achieved by calling Matlab Scriptnode. Then, the accuracy of the built system in detection of corn quality was verifiedthrough sample experiment and proposed some improved measuresThe work in this article can realize objective, rapid and accurate analysis of cornquality and offer some leading effect on applying machine vision technology to detectionof maize quality.