【Objective】Aiming at the problems of highway pavement distresses, such as the existence of many categories, large scale differences and high background complexity, a highway pavement distress detection algorithm with improved YOLOv7 is proposed.【Method】Firstly, the display visual center module EVC is introduced into the neck network to fully obtain the global and local information of the input features and improve the feature extraction ability for small targets; secondly, the feature fusion module RFECSP is designed to enhance the feature fusion ability for multi-class and multi-scale lesions, and to solve the problem of the loss of detail information and the influence of irrelevant regions that lead to the low detection accuracy; finally, the MPDIoU loss function is used to improve the network. MPDIoU loss function to improve the convergence speed and detection accuracy of the network.【Result】The experimental results show that the algorithm in this paper achieves good results on the RDD 2020 dataset, improves the average detection accuracy by 3.13% compared with the YOLOv7 algorithm, and outperforms the algorithms such as SSD, YOLOv4, YOLOv5, etc.【Conclusion】The algorithm in this paper has a good effect of detecting pavement diseases, and is able to satisfy the requirements of detecting different types of diseases on highway pavements.