Comparative Study of Fire Detection Using SqueezeNet and VGG for Enhanced Performance
A. Akilandeswari, Muhammad Amanullah, S. Nanthini, R. Sivabalan, T. Thirumalaikumari, A Rajasekaran
Abstract
The fire detection algorithm effectively employs deep learning methodologies for accurate and efficient fire recognition. It leverages two prominent Convolutional Neural Network (CNN) architectures, SqueezeNet and VGG16, for feature extraction from fire and non-fire images. The images are preprocessed and fed into these networks to extract deep features. Enhancing the model's interpretability and reducing dimensionality is achieved through Dual Regularized Regression (DRR) feature selection, which combines L1 and L2 regularization. This comprehensive pipeline, integrating deep learning, feature selection and machine learning techniques result in a robust system with notable accuracy rates, achieving, 93.71 % with VGG16 and 90.85% with SqueezeNet.