Forest Fire Detection Model Using Dense Net Architecture

Abstract

Forest and land fires in Indonesia are frequent events and cause significant losses in the health, ecological and social sectors. Human and natural factors play a role in triggering these fires. However, handling forest and land fires still faces obstacles in accurately predicting the location of hot spots, making optimal control difficult. Therefore, it is necessary to develop an intelligent system to detect forest and land fires more effectively. This research aims to create a model that is capable of detecting forest and land fires using a transfer learning approach, utilizing the DenseNet201 architecture to increase detection accuracy. The dataset used in this research comes from the Fire Forest Dataset on the Kaggle site. The feature extraction process was carried out using the DenseNet201 architecture, and the resulting model was tested using the confusion matrix method to classify images into two classes, namely fire and non-fire classes. Through training using the DenseNet201 architecture, an effective model was obtained in detecting forest and land fires. Test results using 380 test data show an accuracy level of 99% in recognizing images of forest and land fires. It is hoped that this research can provide a basis for the development of smart systems that are more sophisticated and effective in overcoming the problem of forest and land fires, as well as protecting the environment and public health in Indonesia.