Transfer learning with dual-stage discrete wavelet transform for enhanced visual malware image compression and classification

Recent advancements in deep learning have revolutionized the visual identification and categorization of malware, shifting from manual feature extraction to leveraging visual representations for efficient information transmission. Transfer learning, in particular, has shown promising results in extracting high-dimensional features from malware images. However, grayscale image classification remains challenging as pre-trained models typically require high-dimensional inputs for optimal performance. To address this gap, this study presents a novel methodology that integrates a Dual-Stage Discrete Wavelet Transform (DWT) with a modified DenseNet-121 pre-trained model for enhanced grayscale malware image analysis. The Dual DWT approach systematically reduces image dimensionality while preserving critical features. It first compressing the images by half, and then further optimizing them after feature extraction. The pre-trained DenseNet-121 model is adapted to process grayscale images of reduced size (112 × 112 × 1), ensuring compatibility without sacrificing essential data. We evaluate the effectiveness of this approach using various machine learning classifiers, with XGBoost demonstrating superior accuracy of 98.58%, precision of 96.12%, recall of 96.03% and F1-score of 96.04%. Our findings not only underscore the viability of modified pre-trained models for grayscale image analysis in malware detection but also open new avenues for efficient multiclassification problem-solving in cybersecurity. Our findings demonstrate the efficacy of combining Dual DWT with transfer learning for grayscale malware analysis, offering a computationally efficient solution for multiclass classification in cybersecurity. This work bridges the gap between pre-trained models and grayscale data, opening new avenues for resource-aware malware detection systems.