基于ShuffleAttention相似目標檢測

——以 SA-YOLOv7為例

打開文本圖片集

中圖分類號：TP391.4；TP183 文獻標識碼：A 文章編號：2096-4706（2025）11-0106-08

Similar Object Detection Based on Shuffle Attention

-Taking SA-YOLOv7 asan Example

REN Yuzhen1，F(xiàn)AN Zhongkui1，F(xiàn)ENG Zhenying2，ZHU Mei1 （1.SchoolofSoftwareEnginering，JangxiUniversityofScienceandTechnology，Nanchang33013，China;2.Guangdong Nikola Energy Technology Co.，Ltd.， Guangzhou 51070o， China）

Abstract： YOLOv7achieves excellent results in object detection，but there is stilla problemofhigh false detectionrate insimilarobjectdetection.ThemainreasonisthatYOLOv7hasinsuffcientabilitytoextractfine-rainedfeatures.Tosolvethe above problems，thisstudyproposesaSA-YOLOv7bectdetectionnetwork.Specificallywithoutchangingtheoverallstructure ofELAN，theatentionmoduleSAismergedwithittofoaSA-ELANmoduletoobtainmorechannelandspatialfeature information，therebyimprovingthedetectionaccuracyofsimilarobjects.The modelconductsalargenumberofcomparative experiments onpublichandand glovesimilarobjectdatasets，explores theinfluenceof thenumberand positionofSAadedto theYOLOv7 network ontheresults.Italsoreveals the underlying principleofSA'srole and deepens the understandingof the Attention Mechanism.The experimental results show that the detection accuracy of SA-YOLOv7 is 7.7% higher than that of YOLOv7， and its mAP@0.5：0.95 is 1.8% higher than that of YOLOv7. Compared with the latest YOLOvl1， it also has a 0.9% （2 detection accuracy advantage.TheresearchonSA-YOLOv7 provides assistance for thedevelopmentofsimilar objectdetection technology.

Keywords： Deep Learning; YOLOv7; Shufle Attention; similar Object Detection

0 引言

隨著深度學習的發(fā)展，機器視覺取得了長足進步，目標檢測作為機器視覺的重要研究方向，涌現(xiàn)出眾多優(yōu)秀算法[，它們在ImageNet、COCO、CIFAR-100等知名數(shù)據(jù)集上取得了優(yōu)異的檢測結果。（剩余12853字）