Theses and Dissertations - UTB/UTPA

Date of Award


Document Type


Degree Name

Master of Science (MS)


Manufacturing Engineering

First Advisor

Dr. Subhash Bose

Second Advisor

Dr. Mounir Ben Ghalia

Third Advisor

Dr. Robert Freeman


A neural network classifier for separating clods from onions during harvesting has been developed. The separator consists of a multi-layer feedforward network that maps textural features computed from gray-scale images of onions and clods into the correct object. Texture features were computed from co-occurrence matrices that specify the spatial relationship of pixel values in an image. The textural features selected for this application consist of homogeneity, contrast, variance, and energy. The network was trained using the back-propagation algorithm. Based on the textural features classification, the effect of changing the network configuration on separation effectiveness was investigated. Factors including network topology and combination of textural feature measures forming the inputs of the network were systematically analyzed. Thirty three different network configurations were evaluated. The best separation effectiveness was obtained with three-layer (3-2-1) network with input set consisting of energy, contrast, and homogeneity feature measures. The separation effectiveness for 3-2-1 network topology was 96 percent. An analysis of integration of the neural network-based vision system with a mechanical separator is presented.


Copyright 2001 Demian Morquin. All Rights Reserved.

Granting Institution

University of Texas-Pan American

Included in

Manufacturing Commons