Design and Evaluation of a Convolutional Neural Network for Early Fire Detection in Static Images

Often caused by human deeds, fire disasters lead to insignificant environmental, social, and financial harm. To reduce these damages, early fire recognition and a self-directed response system are essential for effective disaster management. This paper presents an early fire recognition structure established on fine-tuned Convolutional Neural Networks (CNNs), which are accomplished in detecting fire in inside and outside surroundings. Smart surveillance cameras can detect various irregular actions, including disasters, medical crunches, and Fires. Among these, fire poses the greatest risk, as fading to handle it in its initial stages can outcome in disastrous consequences. Identifying the prospective of CNNs, we developed a lightweight, CNNs-based fire detection system to enhance early warning capabilities. Given the model’s high accuracy, our system can help disaster management teams respond rapidly to fire incidents, aiding in preventing large-scale damage. We trained our model using three different datasets and accomplished impressive results. With Dataset-1 our main dataset, the model touched a great training accuracy of 98%, With a low loss of 0.04%. The validation accuracy was 97%, and the validation loss was 0.05%. On the test dataset, the model performed remarkably well, achieving 97% accuracy with a loss of 0.05%. Experimental results confirm the effectiveness of our approach.