A Neural Network Model for Classifying Duck Eggs Using Computer Vision
|
 |
|
Post a Comment
|
|
|
|
|
CONTRIBUTORS:
|
|
|
PROCEEDINGS TITLE:
|
|
|
YEAR:
|
2007
|
|
PUB TYPE:
|
Conference Paper in Proceedings
|
|
PAGES:
|
55 -
n/a
|
|
SUBJECT(S):
|
balut; duck egg embryo; egg grading; neural network; computer vision
|
|
DISCIPLINE:
|
Engineering and Applied Sciences
|
|
HTTP:
|
http://www.ics.uplb.edu.ph/node/254
|
|
LANGUAGE:
|
English
|
|
PUB ID:
|
103-444-131
(Last edited on
2008/07/19 02:31:05 GMT-6)
|
|
SPONSOR(S):
|
|
|
ABSTRACT:
The presence of pinholes, stretch marks, strain, discoloration, as well as dirt, and their combinations on egg shell surface makes classifying eggs a complex process. In the backyard industry of Balut (duck egg embryo) making in the Philippines, the manual method of egg grading through candling is labor intensive and vulnerable to some mechanical intrusion as human contact with the eggs is unavoidable. Human graders, although through years of experience have become experts of the process, are susceptible to errors because of subjectivity, eye strain, boredom, and tiredness. We propose a color image analysis with a neural network model to automate the grading of eggs using off-the-shelf computer hardware. We acquired, using a digital camera, 750 images of eggs from two balut farms in Quezon province. Fifty percent of the eggs were graded acceptable (positive data) for balut making, and the remaining 50% were graded as rejects (negative data) by the farms’ in-house experts. We used the red, green, and blue (RGB) spectral patterns of the digital images as data sets for a three-layer neural network model (NNM) that will determine whether an egg is accepted or rejected for balut production, given the RGB spectral patterns of the egg’s digital image. We divided the dataset into a training set (70%), a test set (15%), and a validation set (15%), each set with equal number of positive and negative data. We trained the NNM via the method of back-propagation of errors using the images from the training set. The images from the test set were used during training to avoid model over-fitting while the images from the validation set were used to validate the NNM. Validation results show 76% classification accuracy with 7% false positives, making the NNM a potent alternative to manual grading.
|
|
|
|
STATISTICS
|
|
Click on # to view
|
|
Citations
|
|
0
|
|
References
|
|
0
|
|
Comments
|
|
0
|
|
Quality
|
|
0/0.00
|
|
Interest
|
|
0/0.00
|
|
View(er)s
|
|
1/103
|
|
|
|
|
|
|
| Prev |
Next |
|