计算机工程与应用2025，Vol.61Issue(10)：203-213,11.DOI:10.3778/j.issn.1002-8331.2405-0393
改进YOLOv7的小目标检测方法
Improved Small Object Detection Method of YOLOv7
摘要
Abstract
Aiming at the challenging problems of scale variation,complex background interference,missed detection,and false detection in the field of small object detection,an improved YOLOv7 small object detection method is proposed.Based on the YOLOv7 object detection framework,a new adaptive feature collection and redistribution module(AFCR)is added,which can effectively fuse multi-scale features,enhance the detection ability of the model for small objects,and enrich the contextual information of output features.By utilizing feature distillation techniques,the student model can learn key feature representations from the teacher model,avoiding the negative impact of semantic differences across stages,thereby significantly improving the generalization and robustness of the model.The experimental results on three publicly available small object detection datasets,CCTSDB,FloW-Img and TinyPerson,show that the proposed method achieves detection accuracies of 96.4%,84.9%and 33.0%,respectively.Compared with the original YOLOv7 method,mAP@0.5 increases by 6.5,3.9 and 2.9 percentage points,respectively.关键词
小目标检测/YOLOv7/知识蒸馏/多尺度特征融合Key words
small object detection/YOLOv7/knowledge distillation/multi-scale feature fusion分类
信息技术与安全科学引用本文复制引用
冯泰梾,张雪松,宋存利,李光宇,金花..改进YOLOv7的小目标检测方法[J].计算机工程与应用,2025,61(10):203-213,11.基金项目
国家自然科学基金(62276042) （62276042）
辽宁省教育厅科学研究项目(LJKZ0486,LJKMZ20220838). （LJKZ0486,LJKMZ20220838）
