Pavement defect detection algorithm SSC-YOLO for fusing multiscale spatial channels in YOLOv8

To address the challenges of poor detection performance and low computational efficiency in existing pavement defect detection models, a novel algorithm, You Only Look Once with Spatial Slicing Convolution (SSC-YOLO), is proposed. This method enhances detection accuracy by integrating multi-scale spatial feature channels into the You Only Look Once version 8 (YOLOv8) framework. Key improvements include the replacement of conventional downsampling and upsampling operations with Spatial Slicing Downsampling (SSD) and Multichannel Fusion Upsampling (MUP), respectively, which more effectively preserve original feature information. A multi-scale convolutional module with large kernels is also incorporated to expand the receptive field and improve model robustness. In addition, the Multi-scale Fusion Convolution (MFConv) mechanism is introduced to construct a parameter-sharing detection head, thereby improving accuracy while simultaneously reducing memory usage and model complexity. Experimental results on the VOC2007 dataset demonstrate substantial performance gains, with improvements of 2.1% in average recall (AR), 2.2% in mean average precision at 50% Intersection over Union (IoU) ( mAP 50 ), 2.9% in mAP 75 , 3.2% in mAP 50–95 , 3.0% in average precision for small objects ( AP S ), 2.5% for medium objects ( AP M ), and 3.7% for large objects ( AP L ).