Res-Former: An Efficient Malware Detection Network Based on ResNet-50 Deep Convolutional Network and Transformer Models

Abstract

Malware detection remains a critical challenge in cybersecurity due to the continuous evolution and obfuscation of malicious code. This paper proposes Res-Former, a novel hybrid deep learning architecture that integrates the strengths of ResNet-50 and Transformer models for efficient and accurate malware classification. The proposed model leverages the local feature extraction and noise robustness capabilities of convolutional neural networks while incorporating the global dependency modeling and self-attention mechanisms of Transformer architectures. In the preprocessing stage, malware executables are converted into grayscale images and enhanced using histogram equalization to improve feature discriminability. The ResNet backbone extracts hierarchical spatial and structural features, which are subsequently refined through the Transformer branch to capture long-range inter-feature relationships. Finally, an ensemble learning classifier employing bagging and majority voting ensures stable and generalized decision-making. Experimental evaluations conducted on the Malimg dataset demonstrate that Res-Former achieves a superior accuracy of 99.46%, outperforming several state-of-the-art methods including EfficientNetB0, DNN, and IMCBL.