Research On Pavement Crack Recognition Algorithm Based On Generative Adversarial Network Preprocessing

As a modern transportation channel,highway is one of the important infrastructure facilities for the normal operation and development of a country.If pavement cracks are not detected and repaired in time,it will affect the performance of the road,reduce the service life of the pavement,and in severe cases,it is more likely to cause harm to daily driving safety.In recent years,deep learning algorithms represented by convolutional neural networks have achieved tremendous development,and have been applied to the automatic recognition and detection of pavement cracks by a large number of researchers.Their recognition efficiency and accuracy are far superior to traditional digital image processing methods represented by threshold segmentation and edge detection.However,in practical application scenarios of the highway maintenance apartment,the pavement background is complex(e.g.shadows,stains,markings,repairs)and the types of cracks are diverse(e.g.fuzzy cracks,light-colored cracks,wet cracks,net cracks).The characteristics of special cracks are very different,and the data of special cracks is difficult to be collected.The crack recognition model trained by traditional convolutional neural network often has poor recognition effect for special cracks represented by fuzzy cracks and light-colored cracks.Moreover,it is easy to cause false recognition of interference such as shadows and stains,which in turn affects the overall recognition effect of the model.Based on this problem,this paper proposes a pavement crack recognition algorithm based on generative adversarial network preprocessing.The main work includes:1.Design and implement a complete automatic recognition system for pavement cracks based on deep learning algorithms,which can be directly applied to crack recognition tasks under complex pavement backgrounds in practical engineering application scenarios of highway maintenance departments to achieve a more accurate pavement crack recognition effect.2.In the data preprocessing module,this paper proposes a pavement crack image preprocessing algorithm based on generative adversarial network to solve the problem that traditional methods are difficult to detect fuzzy cracks and light-colored cracks.Based on the idea of adversarial learning,the algorithm enhances fuzzy cracks into clear cracks and reconstructs light-colored cracks into dark cracks,retaining high-resolution crack features while suppressing the influence of background noise.It can improve the training and recognition effect of subsequent models.3.In the model training module,this paper proposes a Residual Dense Attention Network for pavement crack recognition.The network is based on the fusion and improvement of the traditional ResNet and DenseNet structures to improve the degradation problem in the training process of deep convolutional neural networks,and it introduces the Convolutional Block Attention Module to further improve the crack feature extraction ability of the network in the actual complex road background.The experimental results show that the pavement crack recognition system designed in this paper has significantly improved the continuity,precision and anti-noise ability of crack detection under the actual complex pavement background.On the actual engineering dataset with precision labels constructed in this paper and multiple public datasets,the system has achieved the optimal recognition effect.