软件导刊2026，Vol.25Issue(3)：141-149,9.DOI:10.11907/rjdk.251618
基于特征融合的宽度—深度学习恶意软件检测
Wide-Deep Learning Malware Detection Based on Feature Fusion
摘要
Abstract
In order to solve the problems of insufficient image representation and feature extraction in existing malware detection methods,a width depth learning malware detection model EffNet IPDBL based on pyramid feature fusion is proposed.The model first visualizes malware as multi-channel images to enhance the characteristics of sample data;Secondly,the dual channel attention mechanism is introduced into Ef-ficientNetB3 module to enhance the ability of channel feature selection and obtain more information to enhance image features;Thirdly,pyra-mid features are used to fuse multi-scale and multi-level image features,while capturing global semantics and local details to improve system robustness;Finally,incremental principal component analysis is used to optimize the pyramid depth width learning module,making it more suitable for processing large-scale data sets.Experiments on self-constructed data sets and Malimg public data sets show that the detection ac-curacy of this model reaches 88.35％and 98.69％respectively,which can effectively detect malware.关键词
恶意软件/宽度学习系统/特征融合/EfficientNet/增量主成分分析Key words
malware/breadth learning system/feature fusion/EfficientNet/incremental principal component analysis分类
信息技术与安全科学引用本文复制引用
赵月爱,刘美晨,展翼乐,王玲..基于特征融合的宽度—深度学习恶意软件检测[J].软件导刊,2026,25(3):141-149,9.基金项目
山西省自然科学研究面上项目(202303021221173) （202303021221173）
山西省科技战略研究专项重点项目(202304031401011) （202304031401011）
