Research on Android malware detection method based on multimodal feature fusion

Existing Android malware detection methods mainly use single-modal data to characterize program features, but fail to fully mine and fuse different feature information, resulting in unsatisfactory detection results. In order to improve the accuracy and robustness of detection, a method for detecting Android malware based on multimodal feature fusion is proposed. Firstly, the permission information is encoded and the Dalvik bytecode data is visualized as a “vector” RGB image. Then, a feedforward neural network and a convolutional neural network are constructed to extract features from the data represented by text and image modalities, respectively. Finally, different weights are assigned to the extracted feature vectors of different modalities, which are added and fused before classification. Experimental results show that the recognition accuracy and F1 score of this method for Android malware both reach 98.66%, and it has good robustness.