In order to solve the problems of low accuracy of the existing road crack detection model and easy to miss and misdetect, a road crack detection model based on MLCA yolov10n was proposed. In view of the slender and irregular shape of road cracks, MLCA (mixed local channel attention) was introduced into the model feature extraction network to enhance the extraction capability of the model for road crack features. Secondly, RepGFPN (reparameterized generalized feature pyramid network) was used to optimise the original neck network to fully integrate the crack feature information at multiple scales. Finally, the Focaler-IoU was used to replace the CIoU loss function and adjust the weights of different crack samples for model training to speed up the convergence. Experimental results on the RDD2022_China dataset show that the improved model improves detection accuracy by 4.4% and mAP(mean average precision) by 2.9% compared to the original YOLOv10n model. Compared with other mainstream target detection models, the improved model demonstrates the best performance in terms of accuracy, recall and computational cost, which validates the effectiveness and superiority of this paper’s method in road crack detection tasks.