RDW-YOLOv8n: a deeply focused algorithm for small object traffic sign detection under complex weather conditions

In autonomous driving technology, accurate detection of traffic signs is crucial for vehicles to perceive their surrounding road conditions. However, the detection of small traffic signs is often affected by changes in lighting and obstruction from obstacles under complex weather conditions, resulting in inaccurate localization and missed detections. To address these challenges, we introduce a new optimized algorithm—RDW-YOLOv8n. We utilize an enhanced small object layer integrated with the ResCBAM module to replace the original large object layer. Unlike traditional methods of introducing small object layers, this method improves detection accuracy by enhancing the network’s perception of small objects through residual attention while effectively addressing the issue of missed detections present in the original algorithm and significantly reducing the number of model parameters. In addition, lightweight depthwise convolution (DWConv) is introduced to further streamline the model without sacrificing accuracy. Finally, the original loss function is replaced with the WIoU loss function to enable the model to focus on the average quality of anchor boxes, balancing the learning of high- and low-quality samples. Experiments conducted on the CCTSDB 2021 dataset show that the RDW-YOLOv8n algorithm achieves a mean average precision (mAP50) of 84.7% and a recall rate (R) of 75.3%, which represent improvements of 6.9 and 5.2 percentage points over the baseline model, respectively. Meanwhile, the model parameter count is reduced by 68.11%, and the F1 score increases by 3.1%. These results effectively demonstrate the detection performance of this algorithm under complex weather. When compared with recent advanced algorithms, RDW-YOLOv8n achieves the highest mAP50 with the smallest number of parameters, highlighting the strong competitiveness and superiority of this algorithm.