基于卷积神经网络的火焰智能识别研究  

Research on Flame Intelligent Recognition Based on Convolutional Neural Network

作　　者：赵阳 刘俊蕾 付旭峰 王征 王海懿 ZHAO Yang;LIU Jun-lei;FU Xu-feng;WANG Zheng;WANG Hai-yi(Songhua River Hydropower Co.,Ltd.,Jilin Fengman Power Plant,Jilin 132000 China)

机构地区：[1]松花江水力发电有限公司吉林丰满发电厂,吉林吉林132000

出　　处：《自动化技术与应用》2023年第11期64-67,共4页Techniques of Automation and Applications

基　　金：国网新源控股有限公司科技项目(SGXYFM00SJJS2100028)。

摘　　要：传统方法无法描述火焰变化特点,导致识别错误率高,识别时间长,为获得更加理想的识别结果,设计基于卷积神经网络的火焰识别算法。采集火焰图像,对火焰图像进行预处理,提高火焰图像的清晰度,并提取火焰识别特征,采用卷积神经网络对特征和火焰状态之间的关系进行模拟,构建火焰智能识别模型。在相同测试平台下,与其他方法进行对照实验,结果表明,所提方法描述火焰的变化特点,大幅度提升火焰识别正确率,同时缩短火焰识别时间,识别整体性能明显优于经典方法,具有较高实际应用价值。Traditional methods can not describe the characteristics of flame change,resulting in high recognition error rate and long recogni-tion time.In order to obtain more ideal recognition results,a flame recognition algorithm based on convolutional neural network is designed in this paper.It collects the flame image,preprocesses the flame image,improves the clarity of the flame image,ex-tracts the flame recognition features,simulates the relationship between the features and the flame state by using convolution neu-ral network,and constructs the flame intelligent recognition model.Under the same test platform,compared with other methods,the results show that this method can not only describe the change characteristics of flame,greatly improve the accuracy of flame recognition,but also shorten the flame recognition time.The recognition performance is obviously better than the classical meth-od,and has high practical application value.

关 键 词：火焰图像 特征向量 卷积神经网络 识别正确率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]