Early Detection of Forest Fire using Deep Image Neural Networks

Forest wealth is the biggest proportionality in the environmental sustainability. The forest wealth is mainly endangered by various natural calamities such as flood, hurricane, fire,volcano, land striding and so on. The forest is also threatened with man made damages such as afforestation, wild-trafficking, animal trade, illegal expansions. The human oriented calamities can be controlled with the existence of severe laws and regulations. The natural calamities are uncontrollable in nature. The only phenomena which can be controlled by human is forest fire. This proposed work provides insights on the detection of the forest fire in early stages through satellite imaging and the same is supplied to the nearest monitoring stations. In these stations we apply deep image neural network models such as InceptionV3, VGG-16, Resnet 50 and DenseNet 201 to determine the presence of forest fire or not through binary classification. The results show that the Densenet 201 has the deepest convolutional layer among all these methods, provides the accuracy of 98.46 % by outperforming all other models.