Study On The Method Of Apple Intelligent Grading Based On Multi-information Fusion

China is a high-yield country of fruit in the world, both the planting area and the output are in the leading position. However, because of the method of post-classification of the domestic fruit which is still remain in the original way, and for the use of the manual initial stage of classification, it results in poor classification accuracy and low efficiency. Adulterated with low-quality fruit in high grade products seriously affects our fruit's competitive position in international market. To solve this problem, method for fruit quality inspection and grading based on machine vision is developed in this paper. The research results can be applied to the fruit grading machine's theory basis and the technical reference.First of all, the apple image's pretreatment methods are studied.By analyzing the noise type of apple image, the LWT de-noising combined with the bivariate shrinkage function (BSF) with local variance estimation is studied, and the research on apple image is carried out. The experimental result shows that this method can reduce noises, and at the same time, can keep the details of the apple image. Compared with the tradition de-noising methods, it can achieve superior de-noising performance.Segment method based on mask picture and edge detecting method are studied. And the massive experiments prove that the segment and edge detecting method are fast and effective. They can also meet the detecting demand of size, shape, surface color and surface defects character of apple image.Secondly, characteristics of apple defects are analyzed in detail, and the method of defect detection is studied.The method of minimum circumcircle and the maximal inscribed circle is proposed to detect the apple's diameter and shape parameter. The experiments show that the apple's shape parameter which is derived from the above method is more consistent with the assessment of people's visual habit.In the space of the HSI color model, the distribution characteristic of each component of apple surface color is studied. It is found that using the threshold segmentation on components of H can achieve a better color parameter.In the space of the HSI color model, the improved watershed algorithm is used for extracting the characteristics of surface defects on the I component of the apple image. The method is more effective on the protruding surface of apple which appears on bumping injury, puncture injury, abrading injury, hail injury and other major defects' detection.Finally, based on multi-information fusion theory, using the improved BP neural network classifier to fuse the size, shape, surface color and surface defects characteristics information of the apple; and to estimate its integrated quality and grade. The experimental results show that the classification accuracy of the trained multi-information fusion neural network classifier is up to 90%. The main reasons for the fault occurred are also analyzed in this study. In a word, the results are acceptable in use of automatic apple grading progress.