GRShellCatcher: A Webshell Detection Framework Based on Opcode Grayscale Mapping and ResNet
Year: 2026
Type: conference-paper
Abstract: Existing webshell detection methods struggle with advanced code obfuscation, fixed-length input constraints, and limited cross-language generalization. To address these issues, we propose GRShellCatcher, a cross-language detection framework targeting dominant server-side paradigms. It extracts underlying opcode sequences from Java and PHP environments, effectively filtering out source-level obfuscation noise. To bypass input length limits and resolve the inherent long-tail distribution of execution flows, these sequences are mapped into two-dimensional grayscale adjacency matrices using an optimized non-linear contrast enhancement strategy. A 50-layer deep residual network, ResNet50, is then employed to reliably extract low-frequency anomalous transitions and global malicious contextual features embedded within these sparse visual representations. Extensive evaluations demonstrate that GRShellCatcher achieves outstanding accuracies of 99.28% and 98.25%, alongside F1 scores of 99.18% and 98.26%, in Java and PHP detection tasks, respectively. By significantly outperforming state-of-the-art baselines, our approach validates the robust practical viability of optimized structural-to-visual conversion for malicious code detection.
Authors: Xiaoxue Liang, Daofeng Li, Yuqin Xie
Institution: Guangxi University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7220651812
