Enhancing Efficiency and Accuracy of Hurricane-Induced Building Damage Detection Using Transfer Learning-Based Image Classification

In the immediate aftermath of a storm, damage assessment is essential for helping disaster management and first responders allocate resources efficiently. The time-consuming and arduous traditional technique of identifying and calculating building damage entails personally examining the damaged regions. In order to improve the effectiveness and precision of building damage identification after a storm, we present a transfer learning-based image classification system in this study. Our strategy intends to accelerate the damage identification process by using the readiness and accessibility of satellite images. We use InceptionnetV3, VGG16, and ResNet50 custom Convolutional Neural Network (CNN) architectures to train and classify photos of damaged and undamaged structures. To evaluate the models' performance in various scenarios, both balanced and unbalanced test sets are used. Experimental findings show how successful our suggested strategy is. VGG16 outperforms the other models in the situation of an imbalanced test set, demonstrating its capacity to manage a variable distribution of damaged and undamaged structures. In contrast, ResNet50 performs better when the test set is balanced, demonstrating its durability in properly assessing damage across many conditions.