Lightweight fire detection algorithm based on improved YOLOv8

Aiming at the limitations of traditional fire detection algorithms in terms of accuracy and real-time detection, a lightweight fire detection algorithm based on improved YOLOv8 is proposed. The Slim-neck is used to improve the Neck network, reduce the number of parameters and computation of the model, and improve the detection performance of the model. The C2f-Star module is designed, and the Star block is introduced to replace the bottleneck structure of the C2f module in the Backbone network, to better capture the information of the image, and further reduce the complexity of the model. The Focaler WIoU boundary loss function is used instead of the original loss function, which reduces the influence of low-quality samples and increases the regressivity of the network bounding box. The experimental results show that the number of parameters and the computational volume of the improved model are reduced by 14.0% and 16.0%, while the precision and the mean average precision are improved by 3.0% and 1.2%, compared with the original model, which can help real-time monitoring and early warning of fire.