Multispectral true temperature inversion algorithm based on QR decomposition

Multispectral radiation temperature measurement technology is extensively applied across military, industrial, and metallurgical sectors. Such as solid rocket engine tail flame and metal surface temperature measurement. To achieve these target material surface temperature measurements. This paper presents a multi-spectral true temperature inversion algorithm that leverages QR decomposition of multi-channel spectral information , designed to enhance the correlation between spectral information extraction and multi-channel data. The algorithm employs matrix QR decomposition to construct a multi-channel spectral information dataset, addressing the issue of inaccurate target temperature measurements attributed to unknown material emissivity. In addition, in order to improve the accuracy of multispectral inversion algorithm, this paper developed a CNN-LSTM-ATTENTION neural network , the convolutional layer and attention mechanism are added to the LSTM neural network, which enhances the correlation of spectral information and is conducive to more in-depth information mining, so as to measure the target real temperature. The algorithm is called QR decomposition and CNN-LSTM-ATTENTION combined true temperature inversion algorithm, abbreviation QRD-CLA, In the simulation experiment, the QRD-CLA algorithm demonstrates a 0.29% increase in accuracy compared to the GIM-LSTM algorithm. Validation through measured data confirms that the QRD-CLA outperforms the GIM-LSTM by 0.4% in terms of accuracy.