The project focuses on developing a model for the automatic detection of cracks on complex surfaces, with particular attention to historic buildings and cultural heritage assets. It is carried out within the framework of the RAISE project and aims to support structural damage assessment, especially in post-earthquake scenarios.

More specifically, a deep learning model based on convolutional neural network architecture has been developed to detect cracks even on decorated or frescoed surfaces, or on surfaces characterized by repetitive patterns, typical contexts for monumental assets but in which traditional methods often perform poorly. 

The input images are divided into small patches and analyzed individually: each patch is classified as either “crack” or “no crack,” and the results are then combined to reconstruct the full image, clearly highlighting the damaged areas.

The images may be acquired by camera- and drone-based surveys, as those designed according to Product 3.3., to support expert assessments during post-seismic building evaluations.

A web application has been developed to allow users to test the model with their own images. The application is available at:

The website is organized into three main sections: an introductory page presenting the project and its objectives, a descriptive section outlining the model’s workflow, and an operational “Try it!” area for image upload. In this section, users can select a photograph and launch automatic classification, obtaining a visualization that highlights the areas of the image potentially affected by cracks.