Classification of eggshell translucent areas for quality determination using alexnet
Loading...
Date
2018-06
Authors
Wong, Hsueh Chung
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Microbial contaminants are usually the biggest problem faced by egg manufacturer.
Bacteria usually penetrate the eggshell through cracks, micro cracks and translucent area are.
A fine-tuned Alexnet and a machine vision system were used for inspection of pattern of
translucent area. Images 150 cracked egg and 30 intact eggs were taken. To fit into the Alexnet,
1010 images were cropped randomly within the boundary of egg in the size of 227×227×3 in
size training and validation data at the ratio of 7:3. The trained network was then evaluated
qualitatively using Google Dream Images and quantitatively by visualizing and prioritizing
channel with highest activation energy in each convolution layer. Extra 100 cropped images
were used as testing images. The trained network was able to detect reject defective egg with
accuracy of 91% with false reject rate of 12.5%. The developed system acted as a preliminary
study for the development of automated quality determination based on translucent area.