Mobile malware detection method using improved GhostNetV2 with image enhancement technique.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Image-based feature extraction and deep learning are crucial for malware detection efficiency.
• Adversarial sample generation techniques significantly challenge current malware detection models.
• Existing models show decreased effectiveness against adversarial samples.

Purpose of the Study:

• To propose an improved GhostNetV2 model for robust malware detection.
• To enhance detection performance for both normal and adversarial malware samples.
• To address the limitations of current deep learning models against adversarial attacks.

Main Methods:

• Android classes.dex files converted to RGB images, enhanced with Local Histogram Equalization.
• Gabor method used for RGB to single-channel image transformation to reduce processing time.
• GhostNetV2 model improved with channel shuffling, efficient channel attention, and optimized activation functions.

Main Results:

• The proposed model achieved 97.7% accuracy for normal malware detection.
• The model attained 92.0% accuracy for adversarial sample detection.
• Outperformed 20 state-of-the-art deep learning models in detection tasks.

Conclusions:

• The improved GhostNetV2 model offers superior performance in malware detection, particularly against adversarial samples.
• The image preprocessing and model enhancements contribute to increased accuracy and efficiency.
• This approach provides a promising solution for robust and efficient malware detection systems.