Few-Shot Fine-Grained Forest Fire Smoke Recognition Based on Metric Learning.

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Current forest fire smoke detection methods use coarse-grained identification, failing to differentiate between fire and non-fire smoke, leading to false alarms.
• Accurate forest fire monitoring necessitates fine-grained smoke identification, which typically requires extensive datasets.

Purpose of the Study:

• To develop a fine-grained smoke recognition method for forest fire detection using limited data.
• To address the challenge of insufficient data in training accurate smoke identification models.

Main Methods:

• Combines fine-grained smoke recognition with few-shot learning techniques.
• Employs metric learning for identifying differences between fire and non-fire smoke.
• Utilizes a novel feature extraction network structure and training methodology.

Main Results:

• Achieved a 93.75% accuracy rate for fire smoke identification.
• Demonstrated good performance in both feature extraction network design and training strategies.
• Successfully identified fire smoke with a limited available database.

Conclusions:

• The proposed few-shot learning approach is effective for fine-grained forest fire smoke detection.
• This method significantly improves accuracy and reduces false alarms in smoke detection systems.
• It offers a viable solution for forest fire monitoring where large datasets are unavailable.