Files in this item
|(no description provided)|
|Title:||A trainable image pattern classification system for detection of damaged soybean seeds|
|Author(s):||Casady, William Walter|
|Doctoral Committee Chair(s):||Paulsen, Marvin R.|
|Department / Program:||Agricultural and Biological Engineering|
|Degree Granting Institution:||University of Illinois at Urbana-Champaign|
Engineering, Electronics and Electrical
|Abstract:||Optical properties of asymptomatic soybean seeds, green seeds, and four classes of soybean seeds discolored by fungi were determined using a spectroradiometer. Knowledge of the optical properties of soybean seeds provided a means of identifying that differences in spectral reflectance existed which could be used for discriminating among asymptomatic soybean seeds and seeds discolored by a pathogen. This knowledge also aided in identifying the feasibility of using color filters as a means for separation of soybean seed classes. Electromagnetic radiation between 436 and 724 nm provided the most information for linear separation of soybean seeds which was useful for identifying differences in the colors of soybean seeds. No single wavelength or combination of wavelengths could be used to linearly separate all classes of soybean seeds. The camera was used without modification by filters to collect data representing soybeans from the entire visible spectrum.
An image pattern classification program was developed to discriminate among asymptomatic soybean seeds, immature seeds and seeds that had been discolored by fungi or a virus. The program was trainable and could be retrained by the user by recording images of exemplars while using a training mode. The algorithm used chromaticity coordinates to correctly classify asymptomatic seeds, seeds infected by C. kikuchii, seeds which belong to a group used by the Federal Grain Inspection Service called "seeds of other colors", and "materially damaged seeds" 94.3%, 97.3%, 85.3%, and 95.9% of the time, respectively. The comprehensive results for all tests yielded a classification accuracy of 93.9% for classification of seeds into classes which conformed to USDA/FGIS grading procedures. The variables used for classification were color, which was expressed using chromaticity coordinates, and seed shape which was estimated using sphericity. The decision function provided a consistent method and a viable alternative for quality inspection of soybean seeds based on color and sphericity. The computer program was easily adaptable to other seeds by retraining.
|Rights Information:||Copyright 1991 Casady, William Walter|
|Date Available in IDEALS:||2011-05-07|
|Identifier in Online Catalog:||AAI9210758|
This item appears in the following Collection(s)
Graduate Dissertations and Theses at Illinois
Graduate Theses and Dissertations at Illinois