To address the problems of insufficient recognition of small-scale cracks, detail loss during downsampling, limited receptive fields, and weak contextual modeling in lightweight road defect detection from UAV imagery, an improved lightweight detection algorithm, EWS-YOLOv8, is proposed. Based on YOLOv8n, SPDConv was introduced into the backbone and neck networks to enhance feature extraction for small targets and alleviate detail loss during downsampling. A C2f_WTConv module was designed to expand the receptive field through wavelet transform and reduce model complexity. EMA was embedded into the backbone network to capture multi-scale contextual information and improve subtle defect perception. Experiments were conducted on a public road defect dataset. Compared with YOLOv8n, the proposed method improves precision and mAP50 by 11% and 3.4%, respectively, while reducing the number of parameters by 24.3% and FLOPs by 21%. The results show that the proposed method effectively improves the detection accuracy of tiny road defects with lower computational cost. It is suitable for real-time deployment in automated road inspection tasks.