改进YOLOv8n模型的火灾场景火焰检测方法

打开文本图片集

Abstract：Aiming at the problem of low accuracy in flame detection caused by complex smoke and dust environments in fire scenes，an efficient and precise flame detection method based on the YOLOv8n model was proposed.First，a variety of fire scene images were selected as the original images for the dataset，and random noise，such as salt and pepper noise， was added to simulate a smoke and dust environment. Second， a median filtering module was embedded at the front of the model's network framework to enhance the network's capability to handle interference noise in smoke and dust environments.Finall，byutilizing Ghost convolution modules and designing crosslayer connection networks at diferent lay levels，the number of parameters was reduced while the generalization capability of the network was optimized.This enable real-time and high-precision flame detection in fire scene with noise interference. Experimental results show that the improved YOLOv8n model had superior real-time performance and detection accuracy performance.

Keywords：flame detection；random noise；YOLOv8n model；median filtering module； lightweight Ghost convolution

火灾的发生对人们的生命和财产造成严重威胁[]，，传统的火灾检测方法采用温度传感器进行检测，这种检测方法造价昂贵且不适用室外场景的应用[2]。（剩余12496字）