The invention relates to a bridge crack detection method based on multi-scale
feature fusion and multi-layer attention, and the method comprises the steps: collecting a crack image set, and dividing the crack image set into a
training set and a
test set; a bridge crack detection
network model is constructed, the bridge crack detection
network model comprises an
encoder and a decoder, and for the
feature extraction stage, the model structure comprises a plurality of convolutional
layers; in the
encoder part, the model takes a Mobilenetv2
network module as a main network and adopts a cavity space
pyramid pooling structure nested with multiple
layers of attention, the decoder part adopts a multi-scale
feature fusion module and a high-low level feature adaptive weighted
feature fusion module structure combined with
Bottleneck Transform, and a prediction map matched with the resolution of an input image is obtained after up-sampling; the problems of
false detection, insufficient precision and poor robustness or generalization of crack image recognition of a traditional
image processing method are solved, and the detection precision and the generalization ability of the model can be effectively improved. And a multi-layer attention cavity space
pyramid pooling structure is nested, so that the extraction effect of crack features is improved.