SY-YOLOv5s: An Improved YOLOv5s Module for Road Vehicle and Pedestrian Target Detection

The continuous increase in road vehicles and growing concerns about traffic safety highlight the critical need for real-time road vehicle and pedestrian detection. Achieving high performance in these tasks, particularly in resource-constrained environments, remains a significant challenge. To address this, we propose SY-YOLOv5s, a model based on YOLOv5s, specifically designed to enhance real-time object detection efficiency and accuracy. Firstly, we replace the C3 module in YOLOv5s with the Spatial and Channel Reconfigurable Convolution (SCConv) module. The key advantage of SCConv lies in that it can concurrently process spatial and channel information within images, facilitating a more nuanced and efficient analysis. Additionally, we incorporate the Triplet Attention module following the Spatial Pyramid Pooling Fast (SPPF) to effectively capture crossdimensional interactions within the input data through its distinctive three-branch structure, which computes attention weights to further augment feature representation capabilities. This strategic combination empowers the network to accurately extract and represent crucial information from images. Experimental results demonstrate that the improved module achieves higher accuracy compared to the original YOLOv5s model, which proves higher efficiency for road vehicle and pedestrian target detection applications.