Automatic monitoring of the bio-colonisation of historical building's facades through convolutional neural networks (CNN)

Built cultural heritage is exposed to various deterioration problems caused by different types of actions. To reduce the need for major interventions, preventive conservation (PC) approaches were proposed, based on data collection, regular monitoring, inspections, and control of environmental factors. Monitoring actions able to depict the evolution of buildings' deterioration state, have been proposed and implemented in real cases. Considering that digital images (DI) of historical facades are constantly collected by different subjects and for different purposes, they represent the widest existing data source to support PC approaches and develop predictive tools. DI of historical fa & ccedil;ades can be used to help in the early recognition of different types of deterioration processes, supporting the creation and application of predictive models based on machine learning (ML) methods. This work proposes a method for the automatic detection of biological colonisation of building facades. A convolutional neural network (CNN) has been trained and tested with images representing the microalgae and cyanobacteria growth process on historical bricks' facades, collected during experimental activities in controlled conditions. The trained model is characterized by an accuracy of 87 % and can recognise bio-colonisation on different types of bricks. The trained model has been applied to a historical building used as a case study. The facades of the case study are constantly monitored by surveillance cameras, and DI of the facades are often collected due to the public function of the building. The study shows that by simply processing these images with the trained network it is possible to detect the first stage of bio-deterioration processes. This work is part of more extensive research for the early detection of different types of building fa & ccedil;ade damages and can be easily implemented where DI coming from surveillance cameras or other sources are available. (c) 2024 Consiglio Nazionale delle Ricerche (CNR). Published by Elsevier Masson SAS. All rights are reserved, including those for text and data mining, AI training, and similar technologies.