ABSTRACT: Forest fires have impacted negatively on ecosystems, cultures, and economies all across the world. Modeling and anticipating the incidence of wild fires are essential to minimize these damages and reducing forest fires because they can help with forest fire prevention strategies. The convolutional neural network (CNN) has emerged as a key state-of-the-art deep learning method in recent years, and its application has enriched a wide range of fields. As a result, we proposed a CNN-based spatial prediction model for forest fire susceptibility. The concept is that this model is used to identify a fire or the beginning of a fire in a forest using (aerial) surveillance data. In the event of a fire, the model might be applied in real time to low-framerate surveillance video or picture and provide a warning. The network will be trained on a dataset that includes images in three categories: 'fire,' 'no fire,' and 'start fire.' The majority of the photographs will be of forests or forest-like situations. Photos labelled 'fire' have visible flames, while images labelled 'start fire' have smoke sensing the beginnings of a fire. Finally, photographs with the title 'no fire' were taken in forests. We will leverage the data augmentation function offered by Keras (Python Deep Learning API) to conduct a series of random transformations on photos before feeding them to the network in order to train a network that generalizes well to new images. Finally, our goal is to create a legible project which handles every aspect of CNN creation and training. Early detection of fire in the forest is very helpful and our biodiversity can be saved.

Keywords: Forest Fire Susceptibility, Convolutional Neural Network Techniques, Machine Learning.

| DOI: 10.17148/IARJSET.2022.9233