Clustering of Post-Disaster Building Damage Levels Using Discrete Wavelet Transform and Principal Component Analysis

Damage assessment of buildings after natural disasters is generally performed manually by a team of experts at the disaster site, making it prone to human error and resulting in low accuracy in classifying the level of damage. This research aims to develop a more efficient and accurate method in post-disaster building damage assessment by integrating Discrete Wavelet Transform (DWT) and Principal Component Analysis (PCA) techniques. The main contribution of this research is the use of DWT as well as the application of this method on more than one image to improve the accuracy of damage level classification. A total of nine unlabelled images of post-disaster buildings were used in this study, which were obtained from the Regional Disaster Management Agency or Badan Penanggulangan Bencana Daerah (BPBD) of Malang City, Indonesia. The methods applied include data pre-processing, DWT decomposition for image analysis to identify features, and clustering using PCA to cluster the level of building damage into light, medium, and heavy categories, which are then evaluated based on accuracy. The results showed that the method yielded 100% accuracy with validation results from surveyors, as evidenced through 2D and 3D visualisations based on principal components (PC1-PC3). These findings confirm that the integration of DWT and PCA can be an effective alternative in improving the accuracy of post-disaster building damage assessment, as well as supporting decision-making in rehabilitation and reconstruction after natural disasters.