面向不平衡样本的物联网轻量级入侵风险识别方法

A Lightweight Intrusion Risk Identification Method for IoT Based on Unbalanced Samples

在不平衡数据的干扰下,物联网的节点负担增大,容易受到轻量级入侵,因此,提出面向不平衡样本的物联网轻量级入侵风险识别方法。基于一维卷积神经网络设计轻量级入侵风险识别模型,为弥补模型训练不够稳定的缺陷,应用DCGAN设计样本数据平衡算法实施样本数据增强处理。将增强后的样本数据作为模型训练中的基础数据,并对构建的轻量级入侵风险识别模型实施训练,实现轻量级的物联网入侵风险识别。测试结果表明,相比剪枝前,剪枝后模型有效参数量的变化幅度达到10倍,在训练中模型很快达到收敛,同时达到了接近1的训练精度,并且本方法的入侵风险识别精度得到了明显的提升,始终稳定在97%左右。

Under the interference of imbalanced data,the node burden of the Internet of Things increases and is prone to lightweight intrusion.Therefore,a lightweight intrusion risk identification method for imbalanced samples in the Internet of Things is proposed.A lightweight intrusion risk identification model is designed based on one-dimensional Convolutional neural network.In order to make up for the lack of stability in model training,DCGAN is used to design a sample data balance algorithm to enhance the sample data.Use the enhanced sample data as the basic data for model training,and train the constructed lightweight intrusion risk identification model to achieve lightweight IoT intrusion risk identification.The test results show that compared to before pruning,the change in the effective parameter quantity of the model after pruning reaches 10 times.During training,the model quickly converges and reaches a training accuracy of nearly 1.Moreover,the intrusion risk identification accuracy of this method has been significantly improved,consistently stable at around 97%.