Files in this item
|(no description provided)|
|Title:||A knowledge-based machine vision system for grain quality inspection|
|Doctoral Committee Chair(s):||Paulsen, Marvin R.|
|Department / Program:||Agricultural and Biological Engineering|
|Degree Granting Institution:||University of Illinois at Urbana-Champaign|
|Subject(s):||Agriculture, Food Science and Technology
|Abstract:||A knowledge-based machine vision system was developed for automatic corn quality inspection. This system consisted of a primitive feature extraction algorithm, several quality-related feature extraction algorithms, and several knowledge-based corn quality inspection algorithms. The feature extraction and corn quality inspection algorithms were developed and their performance evaluated.
The primitive feature extraction algorithm was developed using on-board hardware-based operations. The primitive features were computed in a processing time of less than one second for one object. The quality-related feature extraction algorithms were developed based on the results of the primitive feature extraction algorithm. A geometric dimension measurement algorithm was evaluated. An average color measurement algorithm was used to separate white and yellow corn varieties. The processing time was about 1.3 seconds.
The knowledge-based quality inspection algorithms were developed by training with pre-classified corn samples using knowledge acquisition algorithms. The pericarp damage inspection algorithm provided a successful classification of 95, 80, and 93% for negligible, minor, and severe damage, respectively. The processing time for the pericarp damage inspection program was about 1.0 to 2.5 seconds.
A Fourier profile-based kernel breakage inspection algorithm had an accuracy of 95% for classifying whole kernels as whole and 96% for classifying broken kernels as broken. The processing time of the breakage inspection program was about 1.5 seconds.
A morphological, curvature/symmetry, profile-based kernel breakage inspection algorithm provided a successful classification of 94 and 95% for whole and broken kernels. The processing time for the classification required about 1.5 seconds from grabbing the live image to the final classification result. The software-based neural network classifier required about 0.2 second of the 1.5 second total time.
The RGB and multispectral image-based color discrimination algorithms were also developed and evaluated by separating the color regions of vitreous endosperm, floury endosperm, germ, and red streak areas on white and on yellow corn. The color discrimination functions provided a successful off-line classification rate from 90 to 100% for the color regions of white and yellow corn kernels. The six-band multispectral images recovered more information about the variation of the spectral reflectance of corn kernels than the standard RGB images.
|Rights Information:||Copyright 1993 Liao, Ke|
|Date Available in IDEALS:||2011-05-07|
|Identifier in Online Catalog:||AAI9329098|
This item appears in the following Collection(s)
Graduate Dissertations and Theses at Illinois
Graduate Theses and Dissertations at Illinois
Dissertations and Theses - Agricultural and Biological Engineering