Lightweight fire detection algorithm based on LSCD-FasterC2f-YOLOv8

In this paper, a lightweight fire detection algorithm based on LSCD-FasterC2f-YOLOv8 is proposed to optimize the real-time and accuracy of traditional target detection models in fire scenarios to meet the requirements of efficient monitoring in complex environments. Addressing YOLOv8’s deployment hurdles on limited devices, it innovates in three ways: 1) Introducing FasterC2f, an efficient C2f alternative, cuts parameters and complexity via dimensional reduction. 2) A lightweight LSCD detection head boosts efficiency and accuracy, leveraging shared convolutions to reduce parameters and enhance feature fusion, ensuring precision in challenging fires. 3) Rigorously comparing multiple IoU loss functions optimizes bounding box regression, enhancing detection without extra computation. The experimental results show that the proposed LSCD-FasterC2f-YOLOv8 algorithm achieves excellent performance in the fire detection task, not only achieving significant lightweight in model volume and computational quantity, but also maintaining even higher detection accuracy comparable to the original YOLOv8.