| 摘 要: 针对现有网络入侵检测技术存在的数据不平衡导致检测准确率不足、实时性差和泛化性能低等问题,对基于ResNet(深度残差网络)的入侵检测模型进行改进。在ResNet的每个Dense(全连接)层后添加自注意力层,形成残差连接,旨在通过捕捉长距离依赖关系增强特征表示能力,同时提升网络的学习能力、灵活性和解释性。使用CIC-IDS-2017数据集对新模型进行验证,结果显示,模型的准确率为97.56%,真正例率为97.46%,误报率为4.00%,损失函数值快速收敛至0.044。本文模型与其他文献模型相比,真正例率平均提升约5.62百分点,准确率平均提升约3.94百分点。 |
| 关键词: 网络入侵检测;深度学习;注意力机制;残差网络 |
|
中图分类号: TP393
文献标识码: A
|
| 基金项目: 浙江省重点研发"领雁"计划项目(2022C01238) |
|
| Residual Network Intrusion Detection Model Based on Attention Mechanism |
|
CHEN Tianxiang1, HE Lili1,2, ZHENG Junhong1,2
|
(1.School of Computer Science and Technology, Zhejiang Sci-Tech University, Hangzhou 310018, China;
2.Zhejiang Provincial Innovat ion Center of Advanced Textile Technology, Shaoxing 312000, China)
|
| Abstract: Aiming at the problems of insufficient detection accuracy, poor real-time performance, and low generalization performance caused by data imbalance in existing network intrusion detection technologies, this paper proposes to improve the intrusion detection model based on ResNet (Deep Residual Network). A self-attention layer is added after each Dense layer (fully connected layer) in ResNet to form residual connections, aiming to enhance feature representation ability by capturing long-distance dependencies, while improving the network ' s learning capability, flexibility, and interpretability. The proposed model is verified on the CIC-IDS-2017 dataset, and the results show that the model's accuracy is 97.56% , the true positive rate is 97.46% , the false alarm rate is 4.00% , and the loss function value quickly converges to 0. 044. Compared with other literature models, the proposed model improves the true positive rate by an average of about 5.62 percentage points, and the accuracy by an average of about 3.94 percentage points. |
| Keywords: network intrusion detection; deep learning; attention mechanism; residual network |
