Abstract

In the face of escalating natural disasters, particularly hurricanes, swift and accurate post-disaster assessment is imperative for effective humanitarian assistance and disaster response (HADR). This study delves into the challenges of employing machine learning and deep learning techniques tailored explicitly for post-hurricane damage detection. Leveraging advanced technologies such as synthetic aperture radar (SAR) and remote sensing, the research proposes an innovative methodology. Deep features are extracted using DenseNet201, and dimensionality reduction through Principal Component Analysis (PCA) is applied. A Support Vector Machine (SVM) classifier discerns damage from satellite images. The model is meticulously developed and rigorously tested on the Hurricane Harvey dataset. Results exhibit a promising accuracy of 97.10, laying a foundation for post-disaster damage identification. The comparison is also done with other classifiers based on performance metrics such as Accuracy, confusion matrix, etc., However, the study acknowledges areas for future exploration, including real-time implementation, scalability, interpretability of deep learning models, integrating human expertise, and addressing biases. Tackling these challenges promises advancements in disaster response, aiding timely and effective actions to mitigate the impact of hurricanes and similar calamities.