Lightweight Convolution Neural Network for Image-Based Malware Classification on Embedded Systems
Agung Fathurrahman, Agus Bejo, Igi Ardiyanto
Abstract
In the application of the internet of things (IoT), the hardware devices used are embedded systems, this causes the IoT system to be very vulnerable to malware attacks because the computing resources in the embedded system do not support running conventional security programs. In this research, we propose a lightweight convolution neural network model for image-based malware classification that can be run on embedded systems such as the Jetson Nano. The average accuracy of the proposed model is 96.22% and memory consumption on the GPU is 498MB, when compared to the existing lightweight CNN model for the malware classification, the proposed model has a better accuracy difference of 5.19% and smaller memory consumption difference of 172MB. The proposed model runs successfully on the Jetson Nano with a memory consumption of 3.1GB without any reduction in accuracy. The results show that the proposed method is suitable for devices in the IoT environment and has high accuracy.