AttentionPoolMobileNeXt: An automated constructiondamage detection model based on a new convolutionalneural network and deep feature engineering models

Türkiye experienced a series of devastating earthquakes, impacting millions of people due to the resultant construction damage in 2023. These events highlighted the urgent need for advanced automated damage detection models to aid in disaster response. This investigation introduces the AttentionPoolMobileNeXt model, a novel solution based on a modified MobileNetV2 architecture, to address this challenge. To rigorously assess the model's effectiveness, we meticulously curated a dataset of construction damage instances categorized into five distinct classes. The application of this dataset to the AttentionPoolMobileNeXt model yielded an accuracy of 97%. Additionally, this work extends its contribution by introducing the AttentionPoolMobileNeXt-based Deep Feature Engineering (DFE) model, which further enhances the system's classification performance and interpretability. The DFE model significantly improved test classification accuracy from 90.17% to 97%, surpassing the baseline model. Collectively, AttentionPoolMobileNeXt and its DFE counterpart advance the state-of-the-art in automated damage detection, providing valuable insights for disaster response and recovery efforts.