Optimized feature selection for tropical wood species recognition using genetic algorithm
An automatic tropical wood species recognition system was developed at the Centre for Artificial Intelligence & Robotics (CAIRO), Universiti Teknologi Malaysia. The system classifies wood species by using texture analysis whereby wood surfaces images are captured and the features are extracted f...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2012
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/34656/1/UswahKhairuddinMFKE2012.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | An automatic tropical wood species recognition system was developed at the Centre for Artificial Intelligence & Robotics (CAIRO), Universiti Teknologi Malaysia. The system classifies wood species by using texture analysis whereby wood surfaces images are captured and the features are extracted from these images which are then used for classifications. The system uses Grey Level Co-occurrence Matrix (GLCM) feature extractor and Back Propagation Neural Network (BPNN) classifier and it can classify 20 wood species. The system performs well with over 90% accuracy. However, when more wood species are added for classification, the accuracy was reduced significantly due to enormous variations among wood. In this thesis, feature selection algorithm by wrapper Genetic Algorithm (GA) was added into the system to overcome features redundancy, making the within class features less discriminatory while increasing the discriminatory features of inter class variations. Basic Grey Level Aura Matrices (BGLAM) and Structural Properties of Pores Distribution (SPPD) feature extractors are used instead of GLCM and the classifiers used are k-Nearest Neighbour and Linear classifiers in Linear Discriminant Analysis (LDA). Results of experiments before and after feature selection for all databases are compared and analysed. The feature selection algorithm shows a considerable improvement in the classification accuracy from 86% to 95%. A new mutation operation in the GA for feature selection is also developed to increase the GA convergence rate while maintaining its level of performance. |
|---|