Refining CNN architecture for forest fire detection: improving accuracy through efficient hyperparameter tuning

Abstract

Forest fire detection is one of the critical challenges in disaster mitigation and environmental management. This research aims to increase the accuracy of forest fire detection through improving the convolutional neural network (CNN) architecture. The main focus of research is on efficient hyperparameter tuning, which includes selecting and optimizing key parameters in CNN architectures such as convolutional layers, kernel size, number of neurons in hidden layers, and learning algorithms. By utilizing grid search techniques and heuristic-based optimization algorithms, the resulting CNN model shows significant improvements in detection accuracy compared to previous approaches. The evaluation was carried out using a pre-processed forest fire image dataset, and the results show that architectural refinement and appropriate hyperparameter tuning can substantially improve model performance. Evaluation results comparing two models, VGG16 and the proposed method, show significant improvements over the proposed method. The proposed method shows better capabilities with an accuracy of 95.31% and a precision of 97.22%. This research contributes to developing a more reliable and efficient forest fire detection system, which is expected to be used in real applications to reduce the impact of forest fires more effectively.