Road crack detection algorithm based on improved YOLOv13n

This dataset and code correspond to the journal paper entitled "Road crack detection algorithm based on improved YOLOv13n". It includes the improved YOLOv13n-DLKA model, training code, test code, and the processed road crack dataset used in the experiments. The proposed YOLOv13n-DLKA algorithm introduces a Deformable Large Kernel Attention (DLKA) module to replace the original DS-C3k2 module, which enhances the adaptability to multi-scale and irregular road cracks. The model achieves 69.8% precision, 66.6% mAP@0.5, and maintains lightweight characteristics (5.4MB, 1.7ms inference), making it suitable for mobile and edge device deployment in road inspection scenarios. The repository contains: 1. Complete training and inference code of the YOLOv13n-DLKA model 2. Model configuration files 3. Experimental dataset (Crack dataset including D00, D10, D20, D44 categories) 4. Evaluation scripts and result analysis files All resources can be used to fully reproduce the experimental results in the paper.