DMC-YOLOv11: an improved pavement damage detection model based on YOLOv11

With the continuous growth of road traffic, pavement surface damage detection has become increasingly critical. To address the limitations of existing algorithms in detection accuracy and computational efficiency under complex scenarios, this study proposes DMC-YOLOv11, an enhanced pavement damage detection model based on the YOLOv11 architecture. First, Depthwise Separable Convolution (DWConv) replaces the standard convolutional layers to eliminate redundant feature maps, thereby enhancing correlation feature extraction while reducing parameters and accelerating inference. Second, a C3K2-MCFN structure is introduced by integrating a Multi-kernel ConvFFN (MCFN) with the C3K2 module, which segments high-dimensional features into multiple parts to capture local dependencies across diverse receptive fields, thus improving feature representation. Finally, a Cross-View Interaction Module (CVIM) is designed to fuse feature maps from the backbone and neck, facilitating cross-view information exchange and generating more discriminative features for the detection head. Experimental results on the RDD2022-China dataset demonstrate that DMC-YOLOv11 outperforms the baseline YOLOv11, achieving 2.1% and 1.6% improvements in [email protected] and F1-score, respectively. The proposed method effectively balances accuracy and efficiency, providing a promising solution for robust pavement damage detection in complex environments.