Page 528 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Smoke and fire detection is a key scenario in this solution. Traditional edge-based AI camera
systems frequently generate false alarms due to environmental complexities. This solution
introduces a two-tier AI verification framework:

• The edge small model performs real-time preliminary detection of smoke or flame
patterns to trigger immediate alerts
• Cloud-based visual large model conducts secondary validation through contextual
analysis, eliminating 92% of false positives.
The system further employs adaptive sampling to prioritize high-risk zones, ensuring
comprehensive anomaly coverage without compromising responsiveness.

Since the system's deployment, it has promptly identified multiple safety risks, including
workers sitting or leaning against hazardous goods and vehicles illegally crossing container
zones. Overall, this has significantly enhanced the port’s safety management capabilities.

This project validates the industrial applicability of visual large models while addressing
adoption barriers. The hybrid architecture saves cloud computing resources greatly compared
to pure large-model solutions, and seamless integration with existing CCTV infrastructure cuts
a lot of deployment costs. It also allows rapid onboarding of new risk detection scenarios,
ensuring scalability across diverse operational environments.

This approach represents a significant advancement in AI-driven industrial safety monitoring,
offering a robust, cost-effective and scalable approach to fire risk management in high-stakes
logistics environments.

Use Case Status:

Currently, the entire fireworks large model project plan has entered the production stage and
is being implemented at multiple ports in China. The plan includes hardware, AI platform,
large model algorithms, and applications. The video AI platform is customized for fireworks
detection scenarios, with collaborative analysis tasks of large and small models. When users
use it, they only need to set template parameters to create a fireworks analysis task, which is
then analyzed by the AI platform and provides results to the fireworks warning application.

Partners

Organization: China Telecom[2]
Role: Provision of cloud platform; collaborative R&D

2�2 Benefits of the use case

The project significantly enhances port infrastructure safety management through AI
innovation technologies, driving technological advancement in industrial sectors. By reducing
implementation costs and increasing system flexibility, it promotes digital transformation of
industries and demonstrates the critical role of intelligent technologies in key infrastructure
development.

By implementing advanced fire detection technologies, the project strengthens the safety and
resilience of logistics infrastructure, effectively mitigating potential production safety risks. This

492