Statistical texture representation for wood defect image classification using local binary pattern variants
Extensive research has been done on the automation of wood defect detection, to improve the quality of wood products, reduce human labour errors, and increase sales and production, for the wood industry. Our study extends previous work on the automated inspection of wood to include Malaysian wood sp...
Saved in:
| 總結: | Extensive research has been done on the automation of wood defect detection, to improve the quality of wood products, reduce human labour errors, and increase sales and production, for the wood industry. Our study extends previous work on the automated inspection of wood to include Malaysian wood species in order to positively impact local wood product industries. Every wood species has its own unique constant pattern characteristic, known as its texture, which we can use to differentiate defects. However, humans have a limited ability to detect such defects. During the inspection process, wood is placed on a conveyor belt in different feeding directions. Each defect has multiple shapes, colour tones, and sizes. Even though wood often has no uniform surface, they usually have a distinctive pattern. The objective of the study is to construct a rotation-invariant uniform feature representation of wood defects based on statistical texture features using a variation of Local Binary Pattern (LBP). Experiments used a wood defect dataset from the UniversitiTeknikal Malaysia Melaka (UTeM) database. The wood defect dataset consists of heavy hardwood samples which are Rubberwood, Kembang Semangkuk (KSK), Merbau, and Meranti, which were collected from several secondary wood product factories located in the Bukit Rambai industrial area, Melaka, Malaysia. More importantly, our research gives new perception using a statistical texture representation of the LBP variants in the classification of wood defect images. The basic and variants of the LBP feature set, which was constructed from stage of feature extraction processes using Basic LBP, Rotation Invariant LBP, Uniform LBP, and Rotation Invariant Uniform LBP to significantly discriminate wood defect classes, were further evaluated using different classifiers such as Artificial Neural Network (ANN), K-Nearest Neighbour (KNN) and J48 Decision Tree (J48). By comparing the classification performance results, the Uniform LBP with ANN classifier has achieved 69.9%, and were found to demonstrate the highest accuracy level in the classification of wood defects across multiple wood species. |
|---|