YOLO-CBS: A Model for Fast Object Detection Based on YOLOv7-Tiny

This paper proposes YOLO-CBS, a lightweight object detection model enhanced through dual-path optimization for feature representation and cross-layer fusion efficiency. Firstly, the CBS module (Conv-BatchNorm-SiLU) replaces conventional CBL structures, leveraging the smooth gradient properties of SiLU to improve training stability in deep networks. Secondly, a Squeeze-and-Excitation (SE) channel attention mechanism is embedded at critical nodes of the feature pyramid, adaptively calibrating channel weights to enhance discriminative multi-scale feature expression. Experiments on the COCO dataset demonstrate that the improved model achieves 56.1% mAP, providing a high-precision lightweight solution for scenarios such as drone inspection and mobile vision analytics.