基于全局与局部特征引导的显著性目标检测网络

打开文本图片集

DOI：10.16652/j.issn.1004-373x.2026.01.019 引用格式：，.基于全局与局部特征引导的显著性目标检测网络[J].现代电子技术，2026，49（1）：124-128.

关键词：全局与局部特征引导；显著性目标检测；深度学习；特征增强；多尺度；特征融合中图分类号：TN911.73-34；TP3 文献标识码：A 文章编号：1004-373X（2026）01-0124-05

Salientobject detectionnetworkbasedon globalandlocal feature guidance

WANGZhengkai，ZHOUYongxia （CollegeofInformationEngineering，ChinaJiliangUniversity，Hangzhou31oo18，China）

Abstract：Salientobjectdetectionaimstoidentifysalientregions withinanimage.However，existing methodsoftenperform porly whenhandling images incomplex scenesormulti-scaleobjects.Theexisting networkfailstolocatetheobjectsaccurately whendealing withcomplexscenes，soaglobalandlocalfeatureenhancement moduleisintroducedbasedonthecharacteristics of humanvisualsystemtohelpthenetwork obtain moreabundantandacurate features.For theissueofsuboptimalperformance in multi-scaleobjectdetection，anatention mechanism basedmulti-scalefeature fusion moduleis proposed.This module strengthensthefusionofmulti-scalefeaturesandextractsdeeperglobalfeatures.Anerorlossweightisdesigned.Thediference betwen theunionand intersectionofthe predicted mapsand groundtruth mapsiscaleulatedandtakenasthe weightforthe loss function.Bycomputingpixel-levelrors，thenetwork'ssensitivitytolocalfeaturesandspatialconsistencyareenhanced. Comparisons on fivepublicdatasets againsttwelve advanced methods showthatthe proposed method performs beteron multiple metrics，demonstrating its superiority and efficiency.

Keywords：globalandlocal featureguidance;salientobjectdetection;deeplearning;featureenhancement；multi-scale; feature fusion

0 引言

显著性目标检测旨在对复杂多样的图像中视觉上最明显的区域，即显著性区域进行识别与分割。（剩余7161字）