Fractal and chaotic map-enhanced grey wolf optimization for robust fire detection in deep convolutional neural...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision

Background:

• Deep convolutional neural networks (CNNs) are crucial for image classification.
• CNN architecture self-design presents significant challenges.
• Optimizing CNN exploration and exploitation is key to improving performance.

Purpose of the Study:

• To introduce a novel method for enhancing CNN architecture self-design.
• To improve the exploration and exploitation capabilities of CNNs.
• To boost the classification accuracy of deep learning models.

Main Methods:

• Leveraging the grey wolf optimizer (GWO) for enhanced search.
• Implementing a multi-scale fractal chaotic map search scheme.
• Integrating GWO and chaotic maps for CNN architecture optimization.

Main Results:

• Achieved a classification accuracy of 87.37% across 95 trials.
• Outperformed 23 state-of-the-art classifiers on nine benchmark datasets.
• Demonstrated superior performance in CNN architecture enhancement.

Conclusions:

• The proposed bio-inspired and chaotic/fractal approach significantly advances CNN architecture.
• This method offers a promising direction for future deep learning research.
• Effective optimization of neural architecture leads to improved classification tasks.