Automated highway pavement crack recognition under complex environment
Zhihua Zhang,
Kun Yan,
Xinxiu Zhang,
Xing Rong,
Dongdong Feng,
Shuwen Yang
Affiliations
Zhihua Zhang
Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou, China; National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou, China; Gansu Provincial Engineering Laboratory for National Geographic State Monitoring, Lanzhou, China; Corresponding author. Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou, China.
Kun Yan
Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou, China; National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou, China; Gansu Provincial Engineering Laboratory for National Geographic State Monitoring, Lanzhou, China
Xinxiu Zhang
Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou, China; National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou, China; Gansu Provincial Engineering Laboratory for National Geographic State Monitoring, Lanzhou, China; Gansu Province Key Laboratory of Highway Network Monitoring, Lanzhou, China
Xing Rong
Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou, China; National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou, China; Gansu Provincial Engineering Laboratory for National Geographic State Monitoring, Lanzhou, China
Dongdong Feng
Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou, China; National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou, China; Gansu Provincial Engineering Laboratory for National Geographic State Monitoring, Lanzhou, China
Shuwen Yang
Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou, China; National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou, China; Gansu Provincial Engineering Laboratory for National Geographic State Monitoring, Lanzhou, China
The pavement is vulnerable to damage from natural disasters, accidents and other human factors, resulting in the formation of cracks. Periodic pavement monitoring can facilitate prompt detection and repair the pavement diseases, thereby minimizing casualties and property losses. Due to the presence of numerous interferences, recognizing highway pavement cracks in complex environments poses a significant challenge. Nevertheless, several computer vision approaches have demonstrated notable success in tackling this issue. We have employed a novel approach for crack recognition utilizing the ResNet34 model with a convolutional block attention module (CBAM), which not only saves parameters and computing power but also ensures seamless integration of the module as a plug-in. Initially, ResNet18, ResNet34, and ResNet50 models were trained by employing transfer learning techniques, with the ResNet34 network being selected as a fundamental model. Subsequently, CBAM was integrated into ResBlock and further training was conducted. Finally, we calculated the precision, average recall on the test set, and the recall of each class. The results demonstrate that by integrating CBAM into the ResNet34 network, the model exhibited improved test accuracy and average recall compared to its previous state. Moreover, our proposed model outperformed all other models in terms of performance. The recall rates for transverse crack, longitudinal crack, map crack, repairing, and pavement marking were 88.8%, 86.8%, 88.5%, 98.3%, and 99.9%, respectively. Our model achieves the highest precision of 92.9% and the highest average recall of 92.5%. However, the effectiveness in detecting mesh cracks was found to be unsatisfactory, despite their significant prevalence. In summary, the proposed model exhibits great potential for crack identification and serves as a crucial foundation for highway maintenance.
