首页|基于多模态特征融合的飞机货舱火警探测技术

基于多模态特征融合的飞机货舱火警探测技术

扫码查看
原文链接
  • 万方数据
  • 维普
针对当前飞机货舱火警误报率高及人工特征提取适应性差的问题，使用一维卷积神经网络，建立了多模态融合火警预测模型，进行特征提取，开展模型的评估与验证，将特征提取与分类进行整合，实现了端到端的火警预测任务，提高了模型的可靠性与准确性。采用双波长烟雾探测技术，探测悬浮颗粒物的索特平均粒径、温度、红外光和蓝光的接收光与发射光功率比值。相较于传统火警探测算法将特征提取和分类分开处理策略，按照无火、阴燃、有火3种类别，进行火警信息的分类预测。结果表明，多模态融合的火警探测模型相对于单模态火警探测模型可以达到更高的探测精度，精度可达0。95以上。
Multimodal feature fusion-based fire detection in aircraft cargo compartments
To solve the problems of high false alarm rate and poor adaptability of artificial feature extraction in aircraft cargo compartment fire alarms,this paper uses a one-dimensional convolutional neural network to establish a multi-modal fusion fire alarm prediction model for feature extraction,evaluation and verification of the model,and integration of feature extraction and classification.Because the particle size of fire smoke particles is often smaller than that of interfering particles,when the two have the same volume concentration,the fire smoke particles have a larger surface area concentration.Therefore,this paper uses a dual-wavelength smoke detection module to obtain the scattering signals of blue and red light of smoke particles and the Sauter mean particle size,to distinguish smoke particles from interfering particles.Besides,this paper uses K-type armored thermocouple to obtain the temperature of the fire scene and characterize the fire alarm at different stages.After that,this paper establishes a multi-modal fusion fire detection model through a one-dimensional convolutional neural network combined with different modal fire information.Through the strategy of weight sharing,the convolutional neural network adopts fewer parameters than the artificial neural network,constructs deeper network layers in an end-to-end manner,solves the problem of poor adaptability of artificial feature extraction,and has natural advantages in deep feature extraction.Through the stacking of the convolution layer and pooling layer,the convolution neural network realizes the deep feature extraction of different modal fire alarm information.The classifier is constructed by Softmax and the full connection layer,and the multi-modal fusion is carried out.According to the three labels of no fire,smolder and open fire,the classification warning is carried out.The experimental results show that the end-to-end multi-modal fusion fire prediction model based on the one-dimensional convolutional neural network can reach more than 0.95 in detection accuracy,which is better than all single-modal fire detection models.The results show that the multi-modal fusion fire detection technology can effectively improve detection accuracy.

safety engineeringmulti-modal fusionone-Dimensional Convolutional Neural Network(1D CNN)dual wavelength smoke detectionaircraft cargofire detection

刘全义、韩冬、艾洪舟、王海斌、胡茂绮

展开 >

中国民用航空飞行学院民航安全工程学院,四川广汉 618307

中国民用航空飞行学院民机火灾科学与安全工程四川省重点实验室,四川广汉 618307

安全工程 多模态融合 一维卷积神经网络 双波长烟雾探测 飞机货舱 火警探测

国家自然科学基金民机火灾科学与安全工程四川省重点实验室重点项目四川省省院省校合作项目国家级大学生创新创业训练计划

U2033206MZ2022JB012022YFSY0048202210624024

2024

10.13637/j.issn.1009-6094.2022.2823

安全与环境学报
北京理工大学 中国环境科学学会 中国职业安全健康协会

安全与环境学报

CSTPCD北大核心
影响因子：0.943
ISSN：1009-6094
年,卷(期)：2024.24(2)
  • 11