摘要
为了解决大多数工程材料真实温度的测量问题,提出了基于BP神经网络的多光谱数据处理方法,分析了随机噪声对真实温度计算结果的影响。结果表明:在没有噪声的情况下,训练过的发射率样本真实温度的识别误差在±30K以内;未训练过的发射率样本真实温度的识别误差在±50K以内。随着随机噪声的增大,网络的识别误差也相应增大,但训练过的样本其网络的识别误差较小。说明加大发射率样本可以提高真实温度的识别精度。
This paper presents a muhi-specteral data processing method based on BP neural networks that gives more accurate temperature measurements for most engineering materials by using the neural network to learn the emissivity-wavelength relationship from a training data set, The model also includes the effects of random noise on the temperature measurements. Experimental results show that the recognition error is ±30K for trained emissivity samples which is better than the recognition error of ±50K for untrained samples. Therefore, the temperature recognition accuracy can be improved by enlarging the number of emissivity samples.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2005年第7期999-1001,共3页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金资助项目(60377037)
哈尔滨工业大学校科学研究基金资助项目(HIT.2002.18)
关键词
神经网络
真实温度
多光谱测温
neural network
true temperature
multi-spectral thermometry
