Detecting Concrete Cracks Through Images Based On Deformable Convolution Neural Network

Cracks in in-service structures are often the most intuitive indicators for assessing the condition of structures.In recent years,with the development of the bridge structure,the conventional bridge crack inspection method has been difficult to meet the requirements of modern detection,and the crack detection technology gradually began to change to intelligent.Because of the high robustness and intelligence,the region-based convolution neural network(R-CNN)has shown incomparable advantages in the field of crack image detection.However,limited by the convolution layer and pooling layer of the conventional convolution neural network,there is no internal mechanism to capture the deformation of the object,so the detection of out-of-plane cracks in the image often occurs false detection or missing detection.To overcome this drawback,a new type of region-based CNN crack detector with deformable modules is proposed in the present study.The core idea of the method is to replace the traditional regular convolution and pooling operation with deformable convolution operation and deformable pooling operation.The idea is implemented on three different regular detectors,namely the Faster R-CNN,region-based fully convolutional networks(R-FCN),and feature pyramid network(FPN)based Faster R-CNN.To examine the advantage of the proposed method,the results obtained from the proposed detector and the corresponding regular detector are compared within 72 sets of training model.The results show that the addition of deformable modules improves the mean average precisions(m APs)of the Faster R-CNN,R-FCN,and FPN-based Faster R-CNN for crack detection.More importantly,adding deformable modules enables these detectors to detect the out-of-plane cracks that are difficult for the regular detector to detect.This study confirmed that the embedding of deformable modules will improve the sensitivity and accuracy of conventional crack detectors to crack detection to a certain extent,and can be applied to structural crack detection in real condition.