| The quality inspection of agricultural seeds is decisive to control seeds quality for elevating yield and quality of products, holding the balance of market price, ensuring the seeds safety during storage and transportation, and protecting the benefit of consumers. Actually, the inspected seeds are very limited in China now. Meanwhile, the inspection methods depending on manual stay low efficiency and couldn't adapt to the inspection in large scale. The application of machine vision technique instead of human vision to quality evaluation of agricultural seeds has obvious superiority. The recent researches in the field of seeds inspection are mostly related to the machine vision technique based on digital image analysis.In China, rice seeds play a key role in grain production. The objective of the research is to lay a solid foundation for auto inspection of rice seeds. This dissertation concentrated on research of principles and methods for nondestructive quality inspection of static single rice seed with visible light image. An optimal machine vision system was set up to meet the needs of rice seeds inspection. Three effective image-processing algorithms were developed to detect external defects of rice seeds, such as incompletely closed glumes, germ and disease. Artificial Neural Nets were applied for variety recognition of rice seeds.The main results of the research are as follows:1. The micro-configuration of defects in rice seeds surface were observed with electronic scanning microscope for the first time, which is very helpful to determine the machine vision system resolution and key features of corresponding defects. Based on the hyper-spectral curves of rice seeds (350-2500nm) and reflectance response of color CCD camera, an optimum light source was selected, which was a 450nm to 650nm panchromatic light with enough radiant intensity. In the light of the test results concerned with the effectsof different incompletely closed glumes, germ and disease on the ratio of germination in various storage periods, the rice seeds were classified into six categories to meet the demand of produce actually. An adjustable color machine vision system was developed and the optimal conditions for acquiring clear digital images of rice seeds were determined.2. Each original image was preprocessed to show seed region while the pixels value of background was zero. And then 23 basic dimension, shape and color features were extracted. The classification ability of single feature was analyzed with Kruskal-Wallis test. Feature distributions were observed by graphs. Then features were selected directly or Principal Components Analysis was used to form data sets for classification. An image information base which included typical images and feature values of each category was established to help the inspection based on the knowledge.3. Germ on panicle, disease and incompletely closed glumes are three characteristics of hybrid rice seed, which are actual causes of poor seed quality. On the basis of the various algorithms contrast, three image-processing algorithms were complemented to detect these external defects quickly and accurately. Combined the features of shape and color, the linear discrimination algorithm achieved an accuracy of 100% for the detection of seeds with germ on panicle. A hue histogram feature was selected to detect diseased seeds, which is more stable than the feature of mean hue. The total error of Parzen Windows method using the hue histogram feature is 4% for the three categories inspection of normal seed, spot diseased seed and severe diseased seed. Using Hough transform, the feature of post-processing images was proved to be a good indicator of incompletely closed glumes. The relevant algorithm was developed and the accuracy of 96% for normal seeds, 92% for seeds with fine fissure and 87% for seeds with unclosed glumes were achieved.4. Four varieties recognition algorithms were complemented for the various typical situations of the rice seeds. For the recognition of varieties wit...
|
PDF Full Text Request