Publication: Colour classification of solid wood panels based on K-means, K-medoids and K-modes clustering algorithm
Loading...
Date
2023-08
Authors
Tan, Jun Wei
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Solid wood panels are one of the most used raw materials in wooden furniture making industry, therefore they are required to classified into similar colour classes to improve the quality of the solid wood panel used for making furniture. However, manual classification by experienced workers can be inconsistent due to different factors such as changes in environmental lighting and worker fatigue. To solve this issue, this project used an automated solid wood panel colour classification system by using Python-based software that performs image classification based on machine learning. First, images of solid wood panels are captured by the machine vision system. To reduce the computational complexity, the image of the solid wood panel is pre processed by cropping the image to reduce its resolution. Next, a selection of images will be made by user from the dataset. This selection process involves choosing 30 images to form a group to make sure the centroid of the cluster can be determined correctly. The number of groups will depend on the batch of wood that needs to be separated, which can be divided into 3, 4, or 5 groups. The images will be used to train the model. The trained model will be used to classify the wood images in the dataset.
The trained model and it performance will be graded by using Silhouette Coefficient. Throughout this project the best method to classify the wood images is by using K means clustering algorithm which achieve Silhouette Coefficient of 0.70820 - 0.82761 for model training and Silhouette Coefficient of 0.46263 - 0.50565 for model testing.