Research on Virus Detection Based on Markov Texture and Efficient Attention Network

Computer viruses are malicious programs that spread through networks and damage computer systems. Modern viruses often appear as covert and harmful malware, causing various threats including system/file damage, data theft, and network attacks. This makes virus detection crucial. Traditional methods relying on feature code matching struggle with unknown viruses, high false alarms, and poor real-time performance. While deep learning has addressed some limitations, challenges remain in dataset annotation, model security, and sample classification. This study extracts byte files from Microsoft's public dataset, converts them into grayscale and Markov images to create a new dataset (80% training, 20% testing), and enhances the VGG19 model by modifying its fully connected layer and integrating the CBAM attention mechanism. Finally, on the Markov image dataset, the accuracy of this model reached 99.50%, the precision reached 99.43%, and the recall rate reached 99.50%. Compared with the original model, all indicators have been improved, and the convergence speed has been significantly improved, reflecting its excellent performance.