Detection and classification of forest fire smoke by deep learning methods using Unmanned Aerial Vehicle in Türkiye

It has been observed that forest fires have increased over the years in the world and in Türkiye. This disaster, which emerged as a result of various events such as climate change, human factors or lightning strikes, caused serious damage to the ecosystem and started to pose great dangers. Direct or indirect destruction not only affects animal and plant ecosystems, but also affects many areas such as climate change, tourism, industry and health in the long run. There are many ways to respond to forest fires. Thanks to today's technologies, early intervention in fires has become easier and more systematic than before. Thanks to deep learning algorithms, the chance of responding to fires increases considerably. In this study, we made classification by estimating whether the fire was an old or new fire, thanks to the images taken from the videos taken on the Unmanned Aerial Vehicle, and an accuracy rate output was obtained with deep learning algorithms. Since the noise values of these pictures taken from motion cameras are high, their loss values are higher than other still cameras. For this reason, by creating a model with deep learning algorithms and increasing the data, 96.97% accuracy value was obtained in our test data. As we can see from the result here, it has been observed that the models created from deep learning algorithms are successful by reaching high accuracy rates. Since the noise values of these pictures taken from motion cameras are high, their loss values are higher than other still cameras. For this reason, by creating a model with deep learning algorithms and increasing the data, 96.97% accuracy value was obtained in our test data. As we can see from the result here, it has been observed that the models created from deep learning algorithms are successful by reaching high accuracy rates. Since the noise values of these pictures taken from motion cameras are high, their loss values are higher than other still cameras. For this reason, by creating a model with deep learning algorithms and increasing the data, 96.97% accuracy value was obtained in our test data. As we can see from the result here, it has been observed that the models created from deep learning algorithms are successful by reaching high accuracy rates.