Forest Fire Detection Using Convolution Neural Networks

Abstract

One of the primary causes of environmental damage is forest fires. These fires are the cause for many social impacts like loss of biodiversity and timber resources, extinction of plants and animals and loss of wildlife habitat. To reduce these problems, we are implementing a model which uses Convolution neural network (CNN). Convolution neural networks are mainly used for image classification. By identifying valuable features, CNN can recognize different objects on images. With the use of Convolution neural network our model aims to verify whether a forest fire is noticeable in a picture. Our network is trained using a dataset that includes images divided into three categories: “fire:”, “no fire”, Images labelled “fire” have fire, and images labelled with “no fire” have no fire. Because Keras and TensorFlow offer high level APIs used for effortlessly developing and training models, we also use them. Compared to other techniques like YOLO (You only look once), SSD (Single Shot MutliBox Detector) this Convolution Neural Network produces higher accuracy than those methods. And further this model could also be applied in real time to low frame rate surveillance video and give alert in case of fire.