Year: 2023
Type: article
Abstract: Massive disaster events that have an impact on infrastructure damage, especially buildings, require the latest tools and technology so that an assessment of the damage that occurs can be carried out quickly and efficiently. Artificial Intelligence (AI), especially with machine learning (ML) and deep learning (DL), is one approach that can be a solution. The study uses satellite imagery data from the xBD dataset repository with a proposed two-stage deep learning model to assess the level of damage to buildings consisting of a building segmentation approach using the PSPNet and UNet models with the ResNet-18 backbone and the ResNet50 model for classification. The flood fill algorithm is inserted between the segmentation and classification stages with the aim of producing better extraction of segmentation results. The deep learning model with PSPNet and ResNet-18 produces an evaluation value and an accuracy value of building damage classification which is slightly better than the previous building segmentation research, with F1 values of 0.8494 and 0.8338 for precision, respectively. Referring to the resulting evaluation value, future research is still very open to developing models to achieve better results.
Institution: Binus University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7217001216