Towel G Ourd Seeds Grading Research Based On Machine Vision

This thesis tries to research on the seed-grading using machine vision. The research puts emphasis on the system design of the grading device, as well as the algorithm analysis of image processing. The content involves fields like mechanical design, computer technology, and statistics.The primary coverage of the thesis includes:1. The design of grading system. The slope rotating disk design is applied in this grading system. The seeds are distributed singly in the interval gap on the edge of the disk while rotating and acquired the grading information. Through the automatic control of the corresponding level discharge port's open and close by the electromagnetic relay, the seeds are made to discharge on the effect of inertia and gravity. The design aims at the rapidity, accurate implement and automatic control of seeds discharging grain by grain.2. The design of color differentiation module. Firstly the RGB color model must be chosen. The next step is to detect the values of R, G and B. The seed will be deemed as the third level if one or more of their values surpass the range of threshold value.3. The gray level co-occurrence matrix (GLCM) is applied in the inspection and grading of seeds. The function used for calculating towel gourd seed's GLCM is been programmed. It can only count the surface of towel gourd's seed with ignorance of the seed's background, and then evaluate the corresponding eigenvalue.4. It is clear through comparative study that pixel spacing of GLCM is four pixels. With the average value of 0°, 45°, 90°and 135°all together 4 directions chosen, it is also the most suitable for towel gourd seeds to select angular second moment, contrast, variance and differmoment as the eigenvalue, and the 32 levels as the gray level.5. By comparing eight kinds of system clustering methods, it can be proved that the semi-partial r-squared is the best clustering method. The total misjudgment rate is 3.10%; while the overall misjudgment rate of stepwise discrimination method is only 0.66%. Finally it comes to the conclusion that linear discriminate is used to determine the level of the seeds.