Comparing Machine Learning Algorithms for Real-Time Forest Fire Detection using Color, Motion, and Shape Features

Forest fires are extremely dangerous to ecosystems, animals, and humans; thus, it is crucial to detect them early on to limit the harm they cause. Presenting a novel framework for the detection of forest fires is the objective of this study. To increase detection speed and accuracy, this framework mixes colour, motion, and form information with machine learning algorithms. To capture the dynamic nature of fire—its ever-changing hues, distorted forms, and unending motion—an iterative method is employed. Colour probabilities are determined during the segmentation stage before going to the verification step. The categorisation stage and subsequent picture processing both make use of colour histograms. These characteristics allow the system to distinguish between flames and other things more effectively. Evaluating the framework's efficacy involved using intersection over union (IoU) ratios and frame-based evaluation. While IoU is used to test the system's ability to localise the fire inside an image, frame-based assessment is used to test the accuracy of fire detection. Tests on the VisiFire dataset proved the framework's dependability with an 89.97% true-positive rate and a 10.03% false-negative rate. And with an average frame rate of 21.70 FPS, the system is fast enough for real-time processing workloads. Considering these results, it seems like the suggested approach could greatly enhance forest fire early warning capabilities. It can also maintain a high degree of accuracy while providing quicker response times. Protecting forest ecosystems, reducing human life danger, and improving disaster management are all possible outcomes of integrating this technology into early warning systems. This framework is a huge step forward in the world of modern fire detection systems.