ABSTRACT
The increasing prevalence of fire incidents in various environments underscores the critical need for efficient and accurate system to detect fire and reduce the harms due to the said. In a proposed method, a novel approach to detect and classify the fire scene using custom designed deep neural networks with a remarkable accuracy exceeding 99%. The proposed custom network model is trained on a widely proven dataset comprising annotated images of fire and non-fire class to ensure resilience and generalization Our methodology involves the extraction of high-level features from input images through multiple convolutional layers, capturing intricate spatial hierarchies essential for discriminating fire scenes. The dataset was meticulously curates to include diverse fire scenarios, varying environmental conditions, and non-fire instances to create a comprehensive learning environment for CNN. The training process involves fine-tuning the parameters to optimize the network performance on the said task.
