Research and Evaluate some Deep Learning Methods to Detect Forest Fire based on Images from Camera

Abstract

Forest fires cause great consequences such as ecosystem imbalance, air quality deterioration, as well as direct impacts on human life. Early detection of a forest fire can help prevent and prevent the impact of this natural disaster and have timely remedial methods. Therefore, early forest fire detection is necessary. To accomplish this, many methods have been proposed and tested. In recent years, methods based on deep learning techniques with image data sources have been interesting and applied diversely because they can achieve optimal efficiency as well as cost savings in actual installation and operation. However, not all models give highly accurate results. In this paper, we study and evaluate four popular deep learning models (Xception, Inception-V3, VGG-19 and ResNet152-V2) that apply to forest fire detection based on images collected from cameras. With each model, we design deep learning networks to detect fires. The models were made on the dataset of 1900 images, including fire and no-fire cases. The experimental results show that all four of the above deep learning models can be applied to forest fire detection with high accuracy. In particular, the model using ResNet152-V2 gives the best results, with a fire detection capacity of 95.53%.