Ensemble Learning Approach for Weapon Recognition Using Images of Wound Patterns: A Forensic Perspective

Abstract

Weapon recognition from wound pattern images plays a vital role in forensic investigations by assisting experts in identifying the probable weapon involved in criminal incidents and supporting evidence-based decision making. This study presents ForensicGraphAI, a blockchain-secured hybrid EfficientNet-GAT ensemble learning framework for automated weapon recognition using wound pattern images. The proposed system integrates deep visual feature extraction through EfficientNet-B0 with structural feature learning using Graph Attention Networks constructed from superpixel-based image graphs. Advanced image preprocessing techniques are employed to enhance wound characteristics and improve feature representation. The extracted CNN and graph-based features are fused to achieve robust classification of multiple wound categories, including knife wounds, revolver wounds, shotgun wounds, screwdriver wounds, automatic rifle wounds, invalid inputs, and images without visible wounds. To ensure data integrity and traceability, a blockchain mechanism is incorporated for secure storage and verification of forensic analysis records. Additionally, an intelligent report generation module provides detailed forensic interpretations of classification outcomes. Experimental evaluation demonstrates the effectiveness of the framework in accurately recognizing weapon-related wound patterns while maintaining secure forensic evidence management.