Forest Fire Detection with Combined SVM and Deep CNN Approach
Bendjillali Ridha Ilyas, Bendelhoum Mohamed Sofiane, A. Tadjeddine, Kamline Miloud, Frioui Kamila, Bahidja Boukenadil
Abstract
Our proposed approach for forest fire detection presents a significant advancement over existing techniques. By integrating SVM-based classification with state-of-the-art deep CNN architectures, specifically VGG16 and ResNet50, we achieve outstanding accuracy. Notably, our method attains a remarkable 97.21% training set accuracy on ResNet-50. This exceptional accuracy enhances early fire prediction and minimizes false alarm rates, contributing significantly to environmental conservation and human life preservation. Moreover, our adept utilization of fine-tuning techniques effectively addresses challenges related to poor generalization and overfitting, thereby further enhancing the overall efficacy of our innovative approach.