Improving the Performance of the YOLOv11 Model for Fire and Smoke Detection Using Hyperparameter Tuning

Abstract

Reducing losses and responding quickly to events depends on early detection of fires. Artificial intelligence has contributed significantly to the development of accurate and reliable detection systems for detecting and classifying the presence of objects. The most prominent shortcomings observed in this study are the possibility of the model encountering difficulties in determining dim lighting or the presence of materials similar to smoke, which affects the model’s performance. In this research study, we created an intelligent model for detecting smoke and fires based on images. The YOLOv11 model is the latest and most advanced deep learning model in object identification applications. In order to determine the true set of hyperparameters to determine the best performance of the model, this information was modified through trial and error and evaluation of different settings including batch size, learning rate, optimizer type, and adding dropout rate. A large database collected from surveillance cameras, internet images and other sources was also used, showing high accuracy results, with the model achieving 98% accuracy. It was found that the improved model was better than the default model. In addition, there was a 73% decrease in false alarms compared to the default model before the improvement. These results highlight the importance of tuning hyperparameters to improve detection accuracy and reduce errors, resulting in a more robust, efficient, and reliable model that can be used to detect smoke and fires in a variety of indoor and outdoor environments.