Fire Classification Using Attention-Based Deep-CNN in Digital Images

Forests play a crucial role in the environment, but they are facing a serious threat from forest fires caused by both natural and human-made factors. To prevent these fires from causing significant damage, an AI-based forest fire detection system is being proposed for use in smart cities to detect and address fires at an early stage. In this paper, we propose a method that uses transfer machine learning techniques to locate fires in digital material. Our approach is predicated on a deep convolutional neural network (CNN) based on MobileV2 network, that was trained using three dataset of images that featured both fire and non-fire related events. The proposed model is capable to recognize visual cues that indicate the presence of fire, and it makes use of the transfer learning techniques to enhance the fire detection accuracy and effectively assesses whether an input image contains fire or no-fire. The experiments are performed on three fire datasets, a combination of different fire datasets to check the performance of our proposed methods and the results are compared with the classical methods as well as state of the art methods using transfer learning techniques. The results show that the proposed approach achieves 98.42% accuracy, 1.58% error rate, 99.47% recall, and 97.42% precision in classifying the fire and no-fire images. Our approach has the potential to be useful in a range of settings, such as the prevention and extinguishment of fires, the conduct of security and surveillance operations, and the maintenance of public.