Abstract: In response to the challenge posed by the difficulty in detecting small road defect areas and the uneven distribution of defect categories, a detection algorithm termed RDD-YOLO was introduced, based on the YOLOv7-tiny architecture. Firstly, the K-means++ algorithm was employed to determine anchor boxes that better conform to target dimensions. Subsequently, in the realm of small object detection, the Quantization Aware RepVGG (QARepVGG) module was utilized within the auxiliary detection branch, thereby enhancing the extraction of shallow features. Concurrently, the Addition and Multiplication Convolutional Block Attention Module (AM-CBAM) was embedded into the three inputs of the neck, effectively suppressing disturbances arising from intricate backgrounds. Furthermore, the feature fusion module (Resblock with Receptive Field Block, Res-RFB) was devised to emulate the expansion of receptive fields in human visual perception, consequently amalgamating information across multiple scales and thereby amplifying representational aptitude. Additionally, a lightweight Small Decoupled Head (S-DeHead) is introduced to elevate the accuracy of detecting diminutive objects. Ultimately, the process of localizing small objects is optimized through the application of the Normalized Wasserstein Distance (NWD) metric, which in turn mitigates the challenge of imbalanced samples. Empirical findings underscore the efficacy of the proposed methodology, manifesting in a notable 6.19% enhancement in mAP50 and a 5.31% elevation in F1-Score, all achieved while upholding a frame rate of 135.26 frames per second. Impressively, these advancements are realized through the incorporation of a mere 0.71M additional parameters and a computational overhead of 1.7 GFLOPs, which can meet the requirements for both accuracy and expeditiousness in road maintenance.