<p>The capacity to detect damage to roads is of critical importance in ensuring traffic safety, facilitating infrastructure maintenance, and fostering the development of smart cities. However, existing detection methods still fall short of optimal accuracy and real-time performance due to complex lighting conditions, variations in road surfaces, and multi-scale targets. The present paper addresses the aforementioned challenges by proposing DBG-YOLO, a multi-module collaborative improvement detection method based on YOLOv8. The objective of this initiative is to address these issues through the utilisation of high-performance computing and real-time processing capabilities. By leveraging multi-module collaborative enhancement, the method enhances the accuracy and real-time performance of road damage detection under complex lighting conditions, varying road surfaces, and multi-scale targets. The following innovations have been identified as being of key significance: First, Introduce the Dysample module to enhance the robustness of the downsampling process, improving the retention of fine-grained crack and small object features; Second, Incorporate a Bidirectional Feature Pyramid Network (BiFPN) during the feature fusion stage to enable efficient cross-level information exchange and multi-scale feature enhancement; Third, Incorporating GSConv lightweight convolutions into the backbone network, combining batch convolutions with pointwise convolutions, effectively reduces computational overhead while maintaining detection accuracy. The experimental results demonstrate that this method achieves a significantly improved detection performance compared to traditional YOLO-based approaches on the RDD2022 public road damage dataset. It exhibits excellent robustness and application potential in complex road environments, effectively meeting the practical demands of intelligent transportation and road maintenance.</p>