Abstract: To address the problem of how to efficiently detect multi-stage attack behavior of IoT botnet from massive network traffic data,an IoT botnet attack detection method based on multi-scale hybrid residual network(IBAD-MHRN)is proposed.Firstly,in order to reduce the calculation parameters of the detection model,a feature selection algorithm based on variance threshold(FS-VT)method is proposed in data preprocessing.Secondly,a data image processing strategy that converts data samples into image samples is adopted to fully tap the potential of the deep learning model.Then,in order to solve the deficiency of the traditional botnet detection model with limited representation ability,a multi-stage attack detection model of IoT botnet based on multi-scale hybrid residual network is proposed.The model integrates the feature information extracted at different scales and depths in a hybrid way,and then eliminates the effect of network degradation caused by network deepening through residual connection.Finally,an IBAD-MHRN method for IoT botnet attack detection is proposed by integrating the above models and algorithms.Experimental results show that the detection accuracy and F1 value of the proposed IBAD-MHRN method reaches 99.8%,and the accuracy and F1 value is improved by 0.14% and 0.36% respectively compared with the better convolutional neural network method,which can effectively and efficiently detect multi-stage attacks of Internet of Things botnets.

Key words: Internet of Things, Botnet, Variance threshold method, Residual network, Multi-stage attacks