Enhancing Wildfire Detection using Transfer Learningbased Convolutional Neural Networks

Fires in forests are the most devastating natural hazards, resulting in substantial economic damages and fatalities across the globe. Millions of hectares are lost on an annual basis and experts caution that atmospheric changes will prompt wildfires to arise more repeatedly and with more intensity in the years ahead. For prompt reaction and mitigation actions, fire occurrences must be identified quickly and accurately. Preventing damage to persons and buildings, that are prone to fire, requires effective early fire detection. Many deep learning models were created to map and identify wild fires, assess their severity and forecast their spread in an effort to minimize these risks. This work presents a method for fire detection using four various pre-trained models: MobilenetV2, Inceptionv3, Resnet50 and VGG 16 to evaluate valuable features from input images and identify fire detection. Experimental results demonstrate Mobilenetv2 accuracy of 85.61%, Inception-v3 achieves an accuracy of 86.10%, Resnet50 obtain accuracy 95.34% and VGG16 accuracy of 93.32%. The results of these investigations demonstrate that, in comparison to other existing fire detection techniques, the suggested method (Resnet50) successfully identifies fire zones and attains good classification performance.