HSC-YOLOv7:An Enhanced YOLOv7 for Small Object Detection
作者单位：School of Mathematical SciencesHarbin Normal University
会议名称：《第43届中国控制会议》
会议日期：1000年
学科分类：08[工学] 080203[工学-机械设计及理论] 0802[工学-机械工程]
关 键 词：YOLOv7 Drone Images Small Object Detection
摘 要：Detection of small-scale objects is an emerging research area with substantial potential for growth and significant practical relevance in the domain of object *** issue of false detections in small object detection persists as a complex research *** tackle the accuracy issues inherent in small object detection,an enhanced YOLOv7 object detection model is *** proposed improvements include:Firstly,the model incorporates a specialized section for detecting small targets to augment the network s capability in identifying small ***,it introduces an enhanced SPD-CBAM(SC) network backbone,which integrates SPD-Conv and CBAM attention mechanisms into the original *** integration aims to boost the recognition ability of effective features for small targets while mitigating the impact of ***,it utilizes SIoU to replace the loss function in the conventional model for optimization,thereby enhancing the network s *** comparisons were conducted on the publicly available VisDrone-2019 dataset,revealing that the enhanced network model exhibited improvements over the original network in both mAP@[.5:.95] and mAP@.5 for the selected dataset *** enhanced model demonstrated superior performance,making it more suitable for the complex small object detection scenarios proposed in the present study.