Image Classification of Forest Fires Using Machine and Deep Learning

With the increasing threat presented by climate change, forest fires are becoming more destructive and frequent. These catastrophes present immense danger to our forest ecosystems, infrastructure, and overall health. To dispatch resources in a timely manner, quick and accurate identification of these fires is crucial. Here, machine and deep learning can be leveraged as a solution. In this research, we investigate the classification abilities of Support Vector Machines (SVM), Xception, ResNetSO, and MobileViT. In addition, two transfer learning approaches making use of Xception and ResNet50 as the backbone networks are explored. The popular DeepFire dataset is utilized for the training and comparison of model performance. After tuning each architecture, our results indicate the transfer learning approach with Xception as the backbone provides the highest accuracy obtaining 99.211%.