Scalable Fire Classification with MobileNetV2-Driven Convolutional Neural Networks

Public safety and the avoidance of extensive damage depend on the prompt and precise identification of fire. The creation of effective and trustworthy fire classification systems is required due to the rising frequency and intensity of wildfires throughout the world. Climate change is making wildfires a more serious concern, therefore finding creative solutions that support the sustainable development goals is imperative. In order to achieve high accuracy and efficiency, this article provide a fire classification model that uses Convolutional Neural Networks (CNNs) with MobileNetV2 architecture. Taking use of MobileNetV2's lightweight design and depth wise separable convolutions, our model is optimized for real-time applications with limited computational resources, like those present in mobile and embedded devices. The proposed model was trained and assessed using a wide range of datasets that included both fire and non-fire photos. The suggested model successfully distinguished between fire and non-fire events, as evidenced by its remarkable 98% classification accuracy. The model's strong feature extraction capabilities and good cross-environment and cross-condition generalization led to this high accuracy. In addition, the model is refined for low-latency inference, which qualifies it for use in real-world scenarios where prompt and accurate fire detection is critical. The potential of the MobileNetV2-based CNN as a workable solution for real-time fire detection in safety-critical applications is highlighted by the combination of high precision and computational economy.