The Discrimination Of Corn Quality Indicators Using Machine Vision Technology

Corn is one of the major grain and industrial crops and grain reserves in our country. We need to evaluate the quality of corn in the processing, saling, storage and other links of corn and its products. Research on corn quality inspection work is of important practical significance on the smooth development of economy and society stability. The current determination with corn quality indicators are mostly based on the requirement of national standards, inspected by human senses. And the use of machine vision technology in corn quality determination could solve arbitration in manual inspection, inefficiency and other shortcomings. In this paper, digital image processing, image analysis and pattern recognition methods were comprehensively discussed. The basic detection theories and methods of quality indicators of corn based on Machine Vision Technology, such as the content of unsound kernels and impurities were researched. The testing system was constructed in line with the requirements, the main research content and conclusions were as follows:(1) The corn quality detection system based on machine vision device was built according to the experiment requirements,and the part of the system was choiced, which was used for RGB color image acquisition.and prepared for the gray-scale transformation, binaryzation, average filtering, image segmentation, morphological opening and closing operation in order to achieve the effect of noise reduction and image enhancement.(2) The image preprocessing, feature parameter extraction and optimization, and the BP neural network training programs was compiled by using MATLAB software based on machine vision. The images of each type of unsound kernels, organic impurities and inorganic impurities were obtained according to the programs, After image preprocessing, extracting parameters of the grain morphology, color and texture feature and analysis showed there were difference between the characteristic parameters among the whole grain kernels and each type of unsound kernels,the impurities and whole grain kernels, especially the significant difference on some characteristic parameters. The characteristic parameters were processed by the methods of principal component analysis and stepwise discriminant analysis to obtain the effect of dimension reduction and screening, and the principal component scores of vector or characteristic parameters were as the input of the model to set up a corresponding detection and recognition model.(3) For the unsound kernels identification, the 7-15-7 three layers BP neural network model was established for its identification. Results showed that the recognition correction rate of the model for the whole grain kernel, injured kernel, spotted kernel, broken kernel, sprouted kernel, heat-damaged kernel and moldy kernel were 95.00 %, 90.00 %, 91.67 %, 90.00 %, 93.33 %, 95.00 % and 96.67 %, respectively. The identification rate was better.(4) For the identification of corn impurities, as the characteristic parameters between the organic impurities, inorganic impurities and the full corn kernel were obviously different, utilizing a single feature threshold method based on decision tree could effectively identify them, and the identification parameters that were the largest contribution to detection were filtered out after stepwise discriminant analysis. The recognition rate was 96.00 % ~ 98.00 % by using single feature threshold method for various types of impurities; the recognition rate was both 100 % by using linear classifier parameters to identify the full corn kernel and the organic impurities, inorganic impurities and the full corn kernel.