Classification of forest and land fire severity levels using convolutional neural network

Forest and land fires have significant negative impacts on the environment, economy, and public health. These fires cause damage to forest ecosystems, resulting in loss of biodiversity, air quality degradation, and climate change. Assessment of areas post-forest and land fires is crucial for measuring the severity level and planning appropriate rehabilitation measures. This research focus to classify the severity levels of forest and land fires based on photo data obtained from field locations in four villages in Jambi Province. The dataset will be trained into a model using Convolutional Neural Network (CNN) with MobileNetV2 architecture. Based on the evaluation results of training the MobileNetV2 model with two image sizes, (224, 224) and (112, 112), using 50 epochs, it is shown that the highest accuracy was obtained from the model with both image sizes, with an accuracy value of 77.7% and the lowest loss value of 0.618. The use of MobileNetV2 architecture model yielded satisfactory results. MobileNetV2 was considered superior in analyzing the classification model performance on the data used, but there is a need for additional field photo data to improve model training.