Deep learning for bridge component classification and damage detection from UAV imagery Available to Purchase

An innovative framework that integrates convolutional neural networks with unmanned aerial vehicles (UAVs) for enhanced bridge component classification and damage detection is proposed. By leveraging high-resolution UAV imagery and advanced machine learning techniques, this approach addresses critical challenges in structural assessment within complex environments. Detailed datasets were collected from bridges in several regions of Vietnam to validate the framework. A fully convolutional network (FCN) was developed, achieving state-of-the-art results. For bridge component classification, the model delivered an impressive overall accuracy of 94.3%, with a mean intersection over union (IoU) of 84.7%, and robust precision (90.4%), recall (92.9%) and F₁ score (91.5%). In damage detection, the FCN achieved a remarkable overall accuracy of 98.7%, with a mean IoU of 88%, precision of 92.2%, recall of 94.5% and F₁ score of respectively. The FCN demonstrated significant potential for transforming bridge maintenance practices by providing a scalable, efficient and accurate solution for infrastructure monitoring. The results underscore its applicability for real-world deployment and potential to guide future research in bridge engineering and intelligent infrastructure management.