Real-time detection and timely treatment of highway cracks are crucially fundamental for vehicle safety.Rapid identification and comparison of cracks is a new method for monitoring the development and changes of geological disasters especially when they induce cracks.Therefore，this study proposed an intelligent recognition method of highway cracks based on semantic segmentation，establishing a model with dataset，neural network，calculation parameters，and evaluation indicators to rapidly identify the cracks.The results show that，firstly，when the neural network Attention U-net built in this study semantically segments highway cracks，the binary cross loss function value and accuracy rate reach 0.008 7 and 0.998 4，respectively.Secondly，compared with traditional algorithms，the semantic segmentation method shows higher accuracy，reliability，and superiority in intelligent recognition of highway cracks，with a Dice similarity coefficient of 0.978.Thirdly，the semantic segmentation method has better robustness and generalization ability to deal with brightness and noise.