Efficient pavement Distress Detection Based on Attention Fusion and Feature Integration

The images in the pavement distress dataset contain complex backgrounds, which makes manual identification more time consuming. In addition, manual identification requires expert experience and knowledge, which is inefficient and expensive. However, the general distress detection framework based on deep learning loses too much surface feature information, which is essential for crack detection. Therefore, we design an attention module that fuses spatial information and channel information and a feature fusion module that is good at integrating surface feature information. Experiments show that our simple method achieves good performance on the pavement distress dataset.