Knowledge Reasoning- and Progressive Distillation-Integrated Detection of Electrical Construction Violations

Area of Science:

• Computer Vision
• Artificial Intelligence
• Construction Safety

Background:

• Detecting worker violations in electric power construction is challenging.
• Existing methods may lack accuracy and efficiency in complex site environments.

Purpose of the Study:

• To develop an innovative method for accurately detecting worker violation behaviors in electric power construction.
• To improve the efficiency and accuracy of safety monitoring systems.

Main Methods:

• Constructed a knowledge graph from electric power construction standards and norms.
• Integrated the knowledge graph with an object-detection model using triplets and an adaptive connection network.
• Implemented a progressive multi-level distillation strategy with two teacher models for knowledge transfer.

Main Results:

• Reduced model size from 14.5 MB to 3.8 MB and FLOPs from 15.8 GFLOPs to 5.9 GFLOPs.
• Achieved an AP50 of 92.4%, a 1.8% improvement over the original model.
• Demonstrated significant reduction in contextual information loss during distillation.

Conclusions:

• The proposed method effectively detects worker violations with improved accuracy and efficiency.
• Offers a novel approach for enhanced safety management and monitoring in construction sites.
• Provides a quantitative basis for the method's superiority in resource-constrained environments.