Classification of binary insect images using fuzzy and gaussian artmap neural networks
Object recognition and classification is an essential routine in our daily lives. Our eyes act as a camera capturing the image of particular object and sending it to the brain to be recognized. Thus, the eye vision system inspires researchers to create machine vision systems. As a significant part...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Subjects: | |
| Online Access: | http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/63439/1/Page%201-24.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/63439/2/Full%20text.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Object recognition and classification is an essential routine in our daily lives. Our eyes act as a camera capturing the image of particular object and sending it to the brain to be recognized. Thus, the eye vision system inspires researchers to create machine vision
systems. As a significant part of the machine vision system, this research focused on two
(2) important phases of the system; feature extraction and classification. As for the
feature extraction six (6) different types of moment invariant techniques namely
Geometric moment invariant (GMI), United moment invariant (UMI), Zernike moment
invariant (ZMI), Legendre moment invariant (LMI), Tchebichefmoment invariant (I'MI)
and Krawtchouk moment invariant (KMI) are used to extract the global shape features
of the binary insect images. These features are then channeled to the Fuzzy ARTMAP
(FAM) and Gaussian ARTMAP( GAM )neural network to be classified and recognized.
In the GAM neural network, a gamma threshold is proposed to find the optimal value
for gamma parameter acting as the initial value for a Gaussian distribution in the
training phase. It is found that KMI is the best technique for features extraction of the
global shape information of the insect images as compared to GMI, UM!, ZM!, LMI and
TMI. The finding is based on the lowest value of Total Min Absolute Error (I'PMAE)
(0.03%-1.01). The training and testing method for both neural networks is based on 4-
folds cross validation technique. It is also found that the performance of F AM neural
network is influenced by the types of normalization technique utilized. The Improved
Linear Scaling (ILS) normalization technique generated the highest classification rate
by the F AM neural network when compared to Unit Range (UR) and Improved Unit
Range (IUR). It is further found that GAM neural network is a better insect
classification technique when compared to F AM neural network producing the
classification accuracy up to 99.58% whereby the classification accuracy of FAM
neural network is 82%. |
|---|