An Automl Approach for the Efficient Classification of Damaged Roofs Using Fuzzy-Transform

The protection of critical infrastructures is essential for the well-being of citizens and the homeland security. Critical infrastructures, e.g., energy and water distribution, transportation, cultural heritage, etc., must be resilient against external threats, for example cyber-attacks or natural disasters. Regarding the latter, the literature has dealt with the classification of buildings damaged by natural disasters such as hurricanes and earthquakes, which can damage roofs and structures. This is especially important in emergency response and recovery operations for damaged structures, and in particular in the cultural heritage domain, where structures must be protected and repaired in a timely manner to avoid major damages to internal artworks. This paper proposes an approach for the classification of damaged roofs after a natural disaster which uses a combination of traditional and novel features based on Fuzzy-Transform, used to feed an AutoML technique named TPOT. The Fuzzy-Transform features are obtained from the application of the Fuzzy-Transform filter on the aerial grayscale image. The features obtained are then given as input to TPOT, which optimizes machine learning pipelines using genetic programming. Our approach outperforms the existing ones available in literature, with an accuracy of 0.965 and an F1-Score of 0.980.