SM-SCAM YOLO: Enhancing Object Detection with Multi-Scale Module and Spatial Channel Attention Mechanism

Abstract

Detecting tiny objects remains a significant hurdle in computer vision, primarily due to scale variation, occlusion, and the loss of detail in low-resolution features. Although YOLO-based detectors are popular for their speed and efficiency in real-time tasks, they often struggle with accurately identifying small objects because of information loss during downsampling. This study introduces an improved YOLO-based model that integrates a Multi- Scale Module (MSM) and a Spatial-Channel Attention Mechanism (SCAM) to address these challenges. The MSM, replacing YOLO's traditional focus layer, captures features at multiple resolutions to enhance localization across various object sizes. Meanwhile, SCAM improves detection accuracy by emphasizing important spatial and channel features, especially in crowded or visually complex scenes. The model's performance was tested on the PKLot dataset, showing notable gains in precision, recall, and mean average precision (mAP) over the standardYOLO-v5, while preserving real-time processing capabilities. This approach offers a practical and scalable solution for tasks like smart parking, traffic surveillance, and automated vehicle monitoring, where detecting small-scale objects is essential.