Evaluation of machine learning methods in detecting buildings collapsed by earthquakes

This thesis study focuses on the analysis of high-resolution unmanned aerial vehicle (UAV) imagery for detecting damaged and collapsed structures in the Hatay Antakya region following the 2023 Kahramanmaraş earthquakes. The study employs object-based image analysis supported by various machine learning (ML) methods to conduct a detailed evaluation. Multi-resolution segmentation (MRS) was applied to Unmanned Aerial Vehicles (UAV) imagery collected in Antakya after the earthquake. Scale, shape, and compactness values were systematically tested to optimize the segmentation parameters. Each parameter combination was subjected to visual inspection and statistical analysis during this process. The spectral, textural, and shape features extracted from the objects were analyzed using Random Forest (RF), Support Vector Machines (SVM), and K-Nearest Neighbors (KNN) algorithms. The overall accuracy (OA), producer's accuracy (PA), and user's accuracy (UA) metrics, used for performance evaluation, provided an objective basis for comparing the algorithms. The results indicate that the K-NN algorithm outperformed the other methods in terms of overall accuracy. The findings of this study highlight the practical benefits of machine learning-based methods in rapid assessment efforts following disasters. This study has increased the success rate of debris detection through the use of Unmanned Aerial Vehicles and highresolution satellite imagery. It is evident that such analyses can be effectively utilized for damage detection in urban areas. However, it has been concluded that overcoming the limitations of the current methods requires the development of more advanced segmentation techniques and the integration of diverse data sources.