Automatic Classification of Reinforced Concrete Bridge Defects Using the Hybrid Network

Early and timely defect detection is very important to maintain the stability of concrete bridges. The current bridge maintenance mainly relies on manual visual inspection, which is labor-intensive, subjective and unreliable. The computer vision method based on deep learning is considered as the state-of-the-art method for structural damage detection owing to its end-to-end training without needing feature engineering. In this article, we propose a novel hybrid network for defect classification of reinforced concrete bridges. The network is composed of convolutional neural network, Transformer and MLP head. The multi-head attention in Transformer block combines knowledge of the same attention pooling via different representation subspaces of queries, keys and values. To evaluate the performance of different models, a comparison between the proposed approach and other state-of-the-art convolutional neural classification networks is conducted. Results indicate that the accuracy, recall and F1_score of our model in the test set are 0.855, 0.850 and 0.852, which is superior to other compared methods.

Automated summary

Early and timely defect detection is crucial for maintaining the stability of concrete bridges. A novel hybrid network, combining deep learning and Transformer module, is proposed for defect classification, achieving superior performance compared to other methods.