Pavement Crack Detection based on Attention Mechanism and Multi Features Fusion

Abstract

As a common pavement disease, pavement cracks pose a potential threat to road safety. Aiming at the crack images with complex background and loud noise, this paper proposes a pavement crack detection network based on attention mechanism and multi-feature fusion. The network adopts an asymmetric encoding-decoding structure. In the encoding stage, the multi-scale extended residual network embedded with dual attention mechanism is used as the feature extractor, which strengthens the network model’s focus on crack pixels and improves the feature extraction ability. In the decoding stage, the feature pyramid module based on spatial attention mechanism is used to fully integrate the semantic information of the deep network with the detailed information of the shallow network. We trained and tested it on three datasets. Results show that the feature extraction accuracy of our algorithm is significantly improved, and it can extract relatively complete cracks in complex background. It is of great significance to pavement maintenance.