Improving Fire Detection Accuracy through Enhanced Convolutional Neural Networks and Contour Techniques.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Fire Safety Engineering

Background:

• Traditional fire detection methods often struggle with small fire instances and complex environmental conditions.
• Existing deep learning models may not adequately address the challenges posed by intricate fire characteristics and varied scenarios.

Purpose of the Study:

• To develop and evaluate a novel fire detection system integrating contour analysis with deep Convolutional Neural Networks (CNNs).
• To enhance fire detection accuracy and robustness, particularly in challenging scenarios involving small fires and complex environments.

Main Methods:

• A new method combining contour analysis for shape detection and deep CNNs for color property analysis was developed.
• A custom labeled dataset was generated, featuring small fire instances and complex scenarios, with selected Regions of Interest (ROIs) for improved model training.
• The enhanced CNN model was trained and evaluated against various metrics.

Main Results:

• The novel approach achieved high performance metrics: 99.4% accuracy, 99.3% precision, 99.4% recall, and 99.5% F1 score.
• The improved CNN model demonstrated superior performance compared to previous CNN models and state-of-the-art methods like Dilated CNNs, Faster R-CNN, and ResNet.
• The method showed significant improvements across all evaluated metrics.

Conclusions:

• The proposed contour analysis and deep CNN method offers a highly effective solution for fire detection.
• This approach demonstrates significant potential for diverse safety and security applications across various settings, including homes, businesses, industrial sites, and outdoor environments.