When deep learning model is applied to road crack detection,there are issues including incomplete crack extraction and sluggish detection speed,an algorithm is proposed for multi-level feature fusion learning of feature maps based on the ResNet34 backbone network and combined with the channel attention and spatial attention mechanisms.The algorithm can produce clean and precise crack segmentation images since it is made up of a feature extraction network and a multi-level feature fusion module.The feature extraction network extracts the hierarchical features of RGB images,the multi-level feature fusion module learns the hierarchical feature information of ResNet34,and the output of each layer is supervised in a hierarchical manner to guide the rapid training of the network.To verify the effectiveness of this network,tests are carried on the public crack dataset,test results show that the algorithm surpasses exiting classical methods in F1,MIoU and FPS.
维普
万方数据