A Peformance Evaluation of YOLOv3 and CIE Lab Color Space Pixel Color Analysis in Fire Detection

Fire is one of the most common disasters that can happen at any time. In order to reduce the likelihood of damage and loss there have been several models that were developed in order to detect these fires and alert authorities so that fire does not spiral out of control. One of these methods is YOLOv3 - a convolutional neural network that can be used to detect fire. Another one of these methods is pixel color analysis which uses the CIE Lab color space in order to detect fire. The models will be tested using low resolution images, and each model is evaluated in terms of precision, accuracy, recall, and F1 measure. This study aims to evaluate the performance of both methods in order to provide a recommendation as to which model is better for detecting low quality fire images. The results of the study showed that pixel color analysis had an accuracy of 93.28%, precision of 97.43%, recall of 84.89%, and an F1-measure of 90.73%. YOLOv3 had an accuracy of 90.02%, precision of 99.77%, recall of 74.39%, and an F1-measure of 85.23%.