Optimizing Forest Fire Detection Systems: A Deep Learning Approach with UAVs and Binary Sand Cat Swarm Optimization

Devastating forest fires are an international environmental crisis that endangers both people and their valuable natural resources. Forest fires are becoming more common and more destructive, leading to significant depletion of natural resources. A combination of human interference with natural processes and the effects of climate alteration have contributed to a recent uptick in the frequency of wildfires. Combating this issue requires firefighters to respond swiftly and identify fires from early smoke so they can prevent the fires from spreading. An efficient system for detecting and monitoring fires is essential for reducing this risk. The purpose of this study is to investigate and assess the state of the art in forest fire detection and monitoring with the use of deep learning algorithms and drones or unmanned aerial vehicles (UAVs). An efficient and economical way to monitor in real-time and detect fires early is to use drones that are outfitted with specialised sensors and cameras. Using the suggested customised Convolutional Neural Network models, the datasets were tested. Deep Q learning (DQL) is used to extract features, which enhances the accuracy of classification. When compared to other transfer learning methods, such as VGG16, Xception, ResNet50, DenseNet121, Inception ResnetV2, besides Inception V3, their outcomes are evaluated. To analyse and solve the optimal parameter selection problem, the authors suggest using Binary Sand Cat Swarm Optimisation (bSCSO) to fine-tune the parameters of the prototypical. The suggested custom VGG16 model outstripped state-of-the-art models by a significant margin (98.14%), which was measured by its ability to generalise to unknown images.