基于特征增强及多层次融合的火灾火焰检测
Fire flame detection based on feature enhancement and multi-level fusion
赵杰 1汪洪法 1吴凯2
作者信息
- 1. 晋中职业技术学院,山西 晋中 030600
- 2. 太原理工大学,山西 太原 030024
- 折叠
摘要
为提升火灾火焰识别检测方法性能,将传统图像处理与神经网络结合,提出1 种基于特征增强及多层次融合的轻量级火灾火焰检测模型.模型利用多种色彩空间转换算法增强火焰特征信息,并设计双阶段多层次特征提取融合结构,配合空间注意力机制对火焰信息由粗到精进行提取;同时,针对火灾火焰特点,引入由浅到深逐步融合的自适应多尺度融合结构,提升对不同阶段火灾目标的检测精度.研究结果表明:本文模型可有效提升火灾火焰的检测效果,且具有更高的稳定性和鲁棒性,可准确高效地实现火灾火焰检测.研究结果可为现有火灾检测设备提供更准确的识别结果,从而更好地预防火灾事故发生.
Abstract
In order to improve the recognition and detection performance of fire flame methods,a lightweight fire flame detec-tion model based on feature enhancement and multi-level fusion was proposed by combining traditional image processing with neural network.Multiple color space conversion algorithms were used in the model to enhance the flame feature information,and a two-stage multi-level feature extraction fusion structure was designed,which was combined with spatial attention mecha-nism to extract the flame information from rough to fine.At the same time,based on the characteristics of fire flame,an adap-tive multi-scale fusion structure that gradually integrated from shallow to deep was introduced to improve the detection accura-cy of fire objects in different stages.The results show that the proposed model can effectively improve the detection effect of fire flame,and has higher stability and robustness,which can accurately and efficiently achieve the fire flame detection.The research results can provide more accurate identification results for existing fire detection equipment,so as to better prevent fire accidents.
关键词
火灾火焰检测/神经网络/特征增强/多层次融合/自适应多尺度Key words
fire flame detection/neural network/feature enhancement/multi-level fusion/adaptive multi-scale引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据