Online exam cheating detection and blockchain trusted deposit based on YOLOv12

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• Online examinations offer flexibility but face challenges from diverse and covert cheating behaviors.
• Current anti-cheating methods lack real-time detection capabilities and reliable evidence management.
• Maintaining the fairness and integrity of online assessments is crucial.

Purpose of the Study:

• To develop an integrated solution for real-time online exam cheating detection.
• To enhance the efficiency and accuracy of cheating detection models.
• To establish a credible and immutable system for preserving evidence of cheating.

Main Methods:

• Lightweight modifications to the YOLOv12 model (YOLOv12NoAttn) by optimizing the backbone and head networks for faster inference.
• Implementation of a trusted evidence depository using Hyperledger Fabric blockchain and IPFS distributed storage.
• Recording detection metadata and content identifiers (CIDs) on the blockchain via smart contracts for evidence integrity.

Main Results:

• The lightweight YOLOv12NoAttn model achieved competitive detection performance with improved efficiency (28% fewer parameters, 13% fewer GFLOPs).
• Experimental results demonstrated slight improvements in mAP50 and Recall.
• Ablation studies confirmed the effectiveness of the model's lightweight optimizations.

Conclusions:

• The proposed integrated system offers an efficient, accurate, and trustworthy solution for online exam cheating.
• This approach significantly contributes to upholding fairness and integrity in online assessment environments.
• The combination of optimized AI models and blockchain technology provides a robust framework for academic integrity.