PVitNet: An Effective Approach for Android Malware Detection Using Pyramid Feature Processing and Vision Transformer

Abstract

This presents a significant challenge for detecting and combating malicious software. Users often grant software permissions unknowingly, exposing their devices to risks such as unauthorized access, file manipulation, and malware propagation. Traditional detection algorithms relying on limited permission-based strategies fall short in addressing this issue. To overcome this, we propose PVitNet (Network based On Pyramid Feature processing and Vision Transformer), an Android malware detection method. PVitNet incorporates pyramid feature processing, attention mechanisms, and an automatic feature extraction tool. By leveraging semantic information from feature pyramid models and learning shared characteristics among similar software, we successfully identify Android malware families. Our experiments on the CICMalDroid 2020 dataset demonstrate the effectiveness of our approach, with a 14.96% increase in accuracy and an F1 score of 98.31%.