Research on Image-Based Vulnerability Detection of Embedded Firmware

With the wide application of the Industrial Internet of Things (IIoT), the security issues of embedded device firmware have become increasingly prominent. Traditional vulnerability detection methods rely on reverse engineering and binary code analysis, which encounter difficulties in cross-architecture adaptation and consume excessive resources. This paper proposes an image-based embedded firmware vulnerability detection method. By converting binary files into byte entropy images and combining multi-instance learning and attention mechanisms, this method directly extracts features from the original byte data and conducts vulnerability detection, avoiding the complex process of disassembly. This method generates firmware byte entropy images, uses convolutional neural networks to extract local features, and aggregates key image blocks through the attention mechanism, ultimately achieving firmware classification and vulnerability identification. Experiments on the dataset show that this method achieves a detection accuracy of 98.78% in a 64-byte window, with an AUC value of 0.99 on the ROC curve, verifying its feasibility.