摘 要
随着网络攻击的日益复杂化，恶意软件检测已成为网络安全领域的关键任务。传统基于特征签名的方法已难以应对快速变异和隐蔽性强的新型恶意软件，而深度神经网络（DNN）凭借其强大的非线性建模能力，在该领域展现出显著优势。本研究旨在通过优化深度神经网络模型结构与训练策略，提升恶意软件检测的准确性和效率。为此，提出了一种融合多尺度特征提取与自适应权重调整机制的改进模型，同时引入对抗训练方法以增强模型对样本变体的鲁棒性。实验采用公开的恶意软件行为数据集进行验证，结果表明所提模型在分类精度、泛化能力和运行效率方面均优于现有主流方法。具体而言，模型在测试集上的检测准确率达到了98.3%，较基准模型提升了约4个百分点，且误报率降低了近30%。此外，针对模型可解释性不足的问题，设计了一种基于注意力机制的可视化分析框架，能够有效定位恶意行为的关键特征。
关键词：恶意软件检测，深度神经网络，多尺度特征提取
ABSTRACT
With the increasing complexity of network attacks, malware detection has become a key task in the field of network security. Traditional eigensignature-based methods have been unable to cope with the rapid mutation and hidden new malware, and deep neural networks (DNNS) show significant advantages in this field with their powerful nonlinear modeling capabilities. This study aims to improve the accuracy and efficiency of malware detection by optimizing the structure and training strategy of deep neural networks. Therefore, an improved model combining multi-scale feature extraction and adaptive weight adjustment mechanism is proposed, and adversarial training method is introduced to enhance the robustness of the model to sample variants. The results show that the proposed model is superior to the existing mainstream methods in terms of classification accuracy, generalization ability and operation efficiency. Specifically, the detection accuracy of the model on the test set reached 98.3%, an improvement of about 4 percentage points over the benchmark model, and the false positive rate was reduced by nearly 30%. In addition, to solve the problem of insufficient explainability of the model, a visual analysis fr amework based on attention mechanism is designed, which can effectively locate the key features of malicious behavior.
Keywords: Malware detection, deep neural network, multi-scale feature extraction
目 录
摘 要 I
ABSTRACT II
第一章 绪论 1
1.1 研究背景及意义 1
1.2 恶意软件检测中模型优化的意义 1
1.3 国内外研究现状分析 2
第二章 深度神经网络模型选择与设计 3
2.1 恶意软件特征提取需求分析 3
2.2 常见深度学习模型对比评估 3
2.3 针对性模型架构设计原则 3
2.4 模型参数优化策略探讨 4
2.5 模型设计的可行性验证 4
第三章 数据集构建与预处理优化 5
3.1 恶意软件数据集特性分析 5
3.2 数据标注方法及其影响因素 5
3.3 数据增强技术在恶意软件检测中的应用 5
3.4 数据预处理对模型性能的影响 6
3.5 构建高效数据集的关键步骤 6
第四章 模型训练与性能优化策略 8
4.1 模型训练过程中的挑战分析 8
4.2 超参数调优方法研究 8
4.3 正则化技术在模型优化中的作用 9
4.4 提升模型泛化能力的策略 9
4.5 性能评估指标体系构建 10
结束语 12
谢 辞 13
参考文献 14
