Multi-scale feature fusion model for bridge appearance defects detection

Although the Faster-RCNN model has obvious advantages in defect recognition, it still cannot overcome the challenging problems such as time consuming, small targets, irregular shapes and strong noise interference in bridge defect detection. To deal with these issues, this paper proposes a novel multi-scale feature fusion model (MFF) for bridge appearance disease detection. Firstly, because the Faster R-CNN model adopts region of interest alignment (ROI) pooling, which omits the edge information of the target area, resulting in some missed detection and inaccurate detection and positioning in bridge defect detection. Therefore, this paper proposes a region of interest alignment (ROI Align) model (MFF-A) based on regional feature aggregation for multi-scale feature fusion, which reduces the missed detection rate of bridge defect detection and improves the positioning accuracy of target area. Secondly, the Faster R-CNN model is insensitive to small targets, irregular shapes, and strong noise in bridge defect detection, which resulting in long training time and low recognition accuracy. So as to a lightweight network and a feature pyramid network structure fusing multi-scale features (MFF-L) are used to shorten the training speed and improve recognition accuracy. Finally, the effectiveness of the proposed method is showed on the bridge disease dataset and the public CFD dataset.