To address the limitations in detection accuracy and inference speed in current road crack detection models, this paper proposes a novel YOLOv8-Crack network model. Based on YOLOv8n, this model incorporates multiple key structural optimizations, including the introduction of the NWD loss function to reduce dependency on aspect ratios of bounding boxes, thus improving detection capability for irregularly shaped cracks. The Slimneck lightweight structure is used to significantly reduce the number of parameters and computational complexity of the model, and accelerate the inference speed. The model also integrates a CA module to enhance the capture of critical feature information. Experimental results on the open-source dataset RDD2022 demonstrate that the YOLOv8-Crack model achieves improvements over the original YOLOv8n, with precision, recall, and mean average precision increased by 1.8%, 3.7%, and 2.6%; respectively, while parameters and computation are reduced by 6.7% and 11.0%.