Improved YOLO Algorithm Based on Multi-Scale Object Detection in Haze Weather Scenarios

Computer vision-based traffic object detection plays a critical role in road traffic safety. Under hazy weather conditions, images captured by road monitoring systems exhibit three main challenges: significant scale variations, abundant background noise, and diverse perspectives. These factors lead to insufficient detection accuracy and limited real-time performance in object detection algorithms. We propose AMC-YOLO an improved YOLOv11-based traffic detection algorithm to address these challenges. In this work, we replace the C3k block's bottleneck module with our novel attention-gate convolution (AGConv), which improves contextual information capture, enhances feature extraction, and reduces computational redundancy. Additionally, we introduce the multi-dilation sharing convolution (MDSC) module to prevent feature information loss during pooling operations, enhancing the model's sensitivity to multi-scale features. We design a lightweight and efficient cross-channel feature fusion module (CCFM) for the path aggregation neck to adaptively adjust feature weights and optimize the model's overall performance. Experimental results demonstrate that AMC-YOLO achieves a 1.1% improvement in mAP@0.5 and a 2.7% increase in mAP@0.5:0.95 compared to YOLOv11n. On graphics processing unit (GPU) hardware, it achieves real-time performance at 376 (FPS) with only 2.6 million parameters, ensuring high-precision traffic detection while meeting deployment requirements on resource-constrained devices.