基于深度学习的自动化软件缺陷检测研究

陈 杰威
广东理工学院

摘要

基于当前深度模型在语义理解与结构建模方面的融合趋势，本文对自动化软件缺陷检测系统进行了研究，阐述了
其系统架构设计、代码语义表示方法、模型结构与训练策略，介绍了检测流程与 CI 环境的集成机制，结合 CodeXGLUE
与 Devign 数据集开展对比实验。研究结果表明，该系统在 F1-score 与检测延时等核心指标上优于现有方法，具备较强的
实用性与工程部署价值。

关键词

深度学习；软件缺陷检测；图神经网络；语义建模

参考

[1]Sharma K L ,Parekh S ,Yadav K A .Observational

constraints using Bayesian statistics and deep learning in f(Q)

gravity[J].Nuclear Physics, Section B,2025,1018117007-117007.

[2]Wang J ,Zheng G .JISS: Joint image super-

resolution and segmentation of magnetic resonance images

via disentangled representation learning[J].Knowledge-Based

Systems,2025,326114057-114057.

[3]Awwal P ,Naval S D .Development of heuristic

adapted serial-based deep learning for efficient adversarial

malware detection framework in windows[J].Knowledge-Based

Systems,2025,326114032-114032.

[4]Singh P H ,Prashar P ,Reddy C B S G , et al.Multimodal

Multi-task deep learning framework for classification of sentiment,

emotion, humor, sarcasm and toxicity from speech[J].Knowledge-

Based Systems,2025,326113995-113995.

[5]Apostol S E ,Pisică G A ,Truică O C .ATESA-

BÆRT: A heterogeneous ensemble learning model for

Aspect-Based Sentiment Analysis[J].Knowledge-Based

Systems,2025,326113987-113987.

[6] 崔宇寅 . 软件自动化测试方法简述与展望 [J]. 电脑知

识与技术 ,2010,6(34):9749-9751+9769.