| Using machine vision technique to detect the corn appearance quality is significant to improve competition of Chinese food-market. After Chinese entrance to the WTO, it becomes more and more important. Static and dynamic detection techniques using machine vision are both studied in this paper. Besides this, two kinds of instruments are developed. One is used in static condition; the other is used in dynamic condition. The contents of the study can be briefly summarized as follows:1. The foundational image-processing algorithm is studied. The image processing algorithms that suitable for corn appearance quality detection are analyzed under static condition and dynamic condition.2. Because of the limitation of image processing algorithm based on rectangular coordinates, a new image-processing algorithm based on polar coordinates is studied. Three regional processing algorithms based on polar coordinates are put forward firstly: the long-short-axis fast detection algorithm, the image processing algorithm of non-uniformity sampling and the given region-processing algorithm. The detection speed of these algorithms is fast and all theses algorithms have rotation invariance.3. A rice appearance-quality detection instrument used in static condition is developed. The instrument can detect several appearance characteristics of rice, such as chalk degree, chalky rice numbers, yellow-colored rice, rice kernel shape, and different kind kernel. Touching kernel image is allowable, so the cost of hardware is reduced and it is much more convenient to users.4. On the basis of morphological erosion, dilation and watershed translation, an improved watershed algorithm is presented, which is adapted to touching kernel image segmentation. The experimental results show that the relative error is less than 2%.5. In terms of the gray statistics and distribution information of kernel region and chalky region distribution characteristic in the kernel region, four algorithms of chalky rice are presented. The chalky rice detection algorithm based on BP network classification is one of the four algorithms and it has the highest recognition accuracy. The accuracy is 99.4%. On the basis of this algorithm, detection algorithm of chalk degree and chalky rice number are put forward. The error of detection algorithm of chalk degree is less than 1% and the error of detection algorithm of chalky rice number is less than 2%.6. In terms of the analysis on the hue histogram, two detection algorithms are put forward firstly: a new detection algorithm based on hue and a new detection algorithm based on BP network classification. The second method solves the detection of the yellow-colored rice effectively and its accuracy of recognition can be reached 96.3%.7. For the detection of the kernel shape, a new detection algorithm based on polar coordinates is presented firstly. This algorithm has many advantages, such as fast detection speed, high accuracy and rotation invariance. It resolves the disadvantage of the detection algorithm based on rectangular coordinates. Compared to the manual method, the maximum absolute error is 0.01 and the maximum relative error is 5.4%. On the basis of the kernel shape detection algorithm, a different-kind-kernelsdetection algorithm is presented. Using this method, the recognition accuracy of japonica rice and sticky rice can be reached 100% and 96%.8. A rice appearance-quality detection instrument used in dynamic condition is developed firstly. It has many advanced functions, such as automatic sample input, dynamic image sampling and analysis, and automatic sample withdraw.9. In terms of dynamic classification of rice, wheat and corn, a new detection algorithm based on BP network classification is presented and the classification result is very good. For rice, when area, kernel shape and chalk rate are chosen to be characteristic parameters, the recognition accuracy of the standard rice, chalky rice, broken rice and different kind kernel can be reached 94%, 92%, 96% and 96%. For wheat,...
|
PDF Full Text Request