Deep Learning-based Malware Detection Using Independent Stream Analysis of RGB and Grayscale Images

As cyberattacks continue to grow in complexity, the need for scalable and accurate malware detection methods has become increasingly urgent. This study presents a deep learning approach for classifying malware based on image representations of executable files. Malware samples were obtained from the VirusShare repository, while benign files were sourced from verified Windows system executables. Both categories were converted into image formats including grayscale, full RGB, and individual red, green, and blue channels using a custom binary-to-image conversion process. A thorough preprocessing procedure was applied, involving data augmentation and standardization, resulting in a large and balanced dataset. A custom convolutional neural network (CNN) was optimized for spatial feature extraction, with class weighting employed during training to address data imbalance. Experimental results demonstrate that the green channel achieved the highest classification accuracy, surpassing other color channels. These findings provide insights for the design of image-based malware detection systems and emphasize the importance of channel-specific analysis in improving classification performance.