.png){kind=icon}
.png){kind=icon}
Return
An Efficient Malicious Sample Identification Scheme Based on Compressed Sensing and Markov Diffusion Model for Sensor Environment
.png){kind=icon}
PRE
AI
DOI：10.1109/JSEN.2025.3624253.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
With the large-scale deployment of sensor devices in the Internet of Things (IoT), the methods and scale of cyberattacks continue to expand, and sensor nodes have become the core target of malware attacks due to their limited computational resources and weak security mechanisms. In this context, traditional malicious sample detection methods show poor robustness and lower accuracy in the face of high-dimensional feature data, sample dynamics, and data scarcity. To improve the efficiency and generalization ability of malicious sample identification in sensor nodes, this article proposes an efficient malicious sample identification scheme based on compressed sensing (CS) and the Markov diffusion model, which effectively alleviates the computational pressure in practical deployment in resource-limited scenarios and improves the identification and determination ability of malicious samples. The scheme constructs a Markov image representation based on a one-order Markov chain model and combines the CS theory to preserve key feature information. Furthermore, we design a Markov diffusion model to generate structurally consistent adversarial samples to enhance model robustness and generalization. Simulation experimental results show that the proposed scheme exceeds the existing methods in terms of classification performance, interpretability, and runtime and efficiency. Test results on the public dataset BIG 2015 demonstrate that the proposed scheme achieves significant performance improvements in classifying nine major malware families (Ramnit, Lollipop, Kelihos_ver3, Vundo, Simda, Tracur, Kelihos_ver1, Obfuscator.ACY, and Gatak), the accuracy and F1-score can reach 97.37% and 0.9746, respectively, the runtime is reduced to 30%, and the model parameter size reduced by 42.6%, which can effectively adapt to resource-constrained and complex threat scenarios in the sensor network.
Keywords:
Diffusion models
Sensors
Data models
Robustness
Computational modeling
Feature extraction
Accuracy
Intelligent sensors
Real-time systems
Adversarial sample generation
compressed sensing (CS)
deep learning classifier
malicious sample identification
Markov chain
Markov diffusion model
Markov diffusion model
Journal
IF:
4.5
Papers:
2.1W
Citations:
7.3W