Empowering fire and smoke detection in smart monitoring through deep learning fusion

Forest desertification might be caused by human activities and climatic changes, which also have catastrophic effects on ecosystems sustaining agriculture and forests. Fire and smoke are significant abnormal occurrence that has the potential to seriously harm both human & animal lives and damage civilization’s property. Considering its demand, it has always needed attention to how effectively can detect fire or smoke from images. The main objective of this work is to develop an accurate fire alert system to avoid fire mishaps. This article describes a prototype Deep Surveillance Unit (DSU) using a method to detect and alert users to the likelihood of forest fires, etc. DSU comprises two detection stages: feature extraction and Classification Module. The features in the extraction block are extracted using Modified deep CNN models like Inceptionv3, MobileNetV2, and ResNet50v2, respectively. The extracted features are passed through the classification module for Multi-class classification. The DeepQuestAI(Fire, Smoke, and Neutral) dataset is used with three classes for experiments. The Inceptionv3, MobileNetV2, and ResNet50v2 attained an accuracy of 87%, 91.33% and 90% respectively. Whereas the proposed fused model attained a higher accuracy of 94.33%. © 2023, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.