Transfer Learning Based Damage Detection in Public Areas

Abstract

The rapidly increasing population and dense urbanization process in cities have made the effective management of public spaces and the sustainability of the infrastructure in these areas important. This process has led city administrations to seek innovative solutions for rapid and accurate detection of damage in public spaces. Traditional damage detection methods are slow and costly, and are insufficient in the face of the dynamic structure of large cities. This situation negatively affects urban security and quality of life. At this point, it is seen that deep learning and artificial intelligence technologies offer a solution to this problem by automating damage detection processes. In this study, an artificial intelligence-based system has been developed for automatic detection of damage in urban public spaces. The MobileNetv2 model was used with its low resource requirement and high success rate. Data augmentation methods were applied to prevent the overfitting problem that may occur due to the limited dataset. The model achieved 83.33%, 84.20%, 83.30% and 83.70% success in terms of accuracy, precision, recall and F1 score, respectively. These findings demonstrate that, the model detects different damage types at a good rate. The results of this study provide an innovative solution in today's rapidly urbanizing world. This solution will provide an effective roadmap to city administrations by quickly and effectively detecting damage to infrastructure elements. This facilitates addressing challenges caused by rapid urbanization. The study carried out in this context has significant value both theoretically and practically.