.png){kind=icon}
.png){kind=icon}
Return
OmBNNet: a resource-efficient FPGA-based obfuscated malware detection method using binarized neural network
.png){kind=icon}
OA
AI
DOI：10.1038/s41598-026-48883-8.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The growing integration of smart technology and artificial intelligence has intensified the demand for strong security measures. Fraudsters continue to upgrade their tactics, including the use of sophisticated obfuscation methods to circumvent traditional defenses. This study examines the influence of obfuscated samples on malware detection systems and examines strategies for improving robustness, as well as the development of an FPGA-based hardware platform for classifying malware with optimal power and resource efficiencies. Initially, binary classification methods utilizing Convolutional Neural Networks (CNN) are analyzed to differentiate between benign and malicious data; these are evaluated using two benchmark datasets: NATICUSdroid and TUANDROMD. Since, computational demands render the direct deployment of CNNs on hardware platforms unsuitable, we developed a computationally efficient model utilizing binarization techniques. This study demonstrates that in addition to accurately classifying input as benign or malicious, it is essential to analyze the influence of obfuscated samples on the detection process. To investigate we obfuscate, application features and develop two new datasets, designated as Obfus_NATICUS and Obfus_TUANDRO. We also propose a lightweight Obfuscation-Resilient Neural Network (OmBNNet) model, trained on both standard and obfuscated features, emphasizing the advantages of integrating such functionalities. OmBNNet model attains an average 10-fold cross-validation accuracy of 96.57% on the Obfus_NATICUS dataset and 99.09% on the Obfus_TUANDRO dataset. The OmBNNet has been successfully embedded into the FPGA SoC (Xilinx ZCU104), achieving post-deployment accuracies of 96% and 98.52% on both datasets respectively, with an average latency of 1.586 ms and a total power consumption of 3.787 watts.
Keywords:
Engineering
Mathematics and computing
Science
Humanities and Social Sciences
multidisciplinary
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3.9
Papers:
27.4W
Citations:
83.5W