Source-based Smoke Characterization and Detection Using YOLOv5

In order to drastically minimize dangers to both life and property in interior situations, early fire detection is extremely important Optimizing YOLOv5 for real-time fire source detection and characterization-specifically, from cooking, cigarettes, and actual fire-is the main goal of this study. A high-resolution device prototype for capturing image, a creation of custom datasets of images related to fires, and an assessment of YOLOv5's accuracy, precision, and speed are among the objectives. By obtaining an average F1-score of 0.94 for cooking and cigarette sources and 0.92 for actual fire, the trained model showed excellent classification accuracy. The results of the confusion matrix show relatively little misclassification, confirming the system's dependability in differentiating between background items and potential fire hazards. Additionally, the model demonstrates its suitability for real-time applications with an average detection speed of 0.547 milliseconds. These results illustrate how YOLOv5 could boost fire detection systems by facilitating immediate and precise fire source identification, lowering false alarm rates, and promoting better safety protocols and risk reduction.