Abstract: The existing pavement crack detection algorithms are easily affected by light and shadow, foreign object obstruction and road marking, which have the problems of missed detection, low detection accuracy and slow detection speed.Proposed a dual-kernel convolutional pavement crack detection algorithm with multi-dimensional feature fusion.Firstly, the triplet attention mechanism is added, cross-dimensional interaction technique is used to construct the interdependence between the channel, height and width dimensions, which endows the model with the ability of multi-dimensional feature fusion; secondly, the dual-kernel convolutional is used to replace the traditional convolutional to realize the model’s lightness; furthermore, the CIoU bounding box regression function is replaced by improved MPDIoU bounding box regression function, which improves the model’s convergence speed.Experiments on YOLOv8 model and RDD2022 dataset show that, compared with YOLOv8n, F1 and mAP are improved by 7.5% and 5.3%,the number of parameters and computation are decreased by 0.2% and 0.7%,respectively, the FPS is improved by 13;compared with SSD,Faster R-CNN,U-Net, and DeepLabv3,F1 is improved by 3.9%,mAP is 2% lower than U-Net, the number of parameters and computation decreased by 0.8% and 3.3% respectively, and the FPS increased by 19.