Midwave FTIR-Based Remote Surface Temperature Estimation Using a Deep Convolutional Neural Network in a Dynamic...

Area of Science:

• Remote sensing
• Deep learning
• Infrared spectroscopy

Background:

• Accurate remote temperature measurement is crucial for analyzing solar effects and environmental conditions.
• Conventional methods struggle with dynamic weather, particularly unknown atmospheric transmissivities, limiting temperature estimation reliability.
• Hyperspectral imaging offers rich spectral information for enhanced remote sensing applications.

Purpose of the Study:

• To develop a novel, real-time remote temperature estimation method using deep learning on midwave infrared hyperspectral images.
• To address the limitations of conventional methods in dynamic weather conditions.
• To demonstrate the practical application of the developed method in assessing thermal stealth properties.

Main Methods:

• A 27-layer surface temperature-deep convolutional neural network (ST-DCNN) regression model was developed.
• Midwave infrared hyperspectral images were acquired using a TELOPS HYPER-CAM MWE over seven months under varying weather conditions.
• The ST-DCNN was trained to predict surface temperatures from 75 spectral channels.

Main Results:

• The proposed ST-DCNN method achieved high accuracy in real-time remote temperature estimation.
• The method demonstrated feasibility and reliability in real-world dynamic weather environments.
• The study successfully utilized the ST-DCNN to evaluate the thermal stealth characteristics of different paints.

Conclusions:

• The novel deep learning approach provides a robust solution for accurate remote temperature measurement, even in challenging atmospheric conditions.
• The ST-DCNN effectively leverages hyperspectral infrared data for precise temperature prediction.
• This method has significant potential for various applications, including environmental monitoring and material science.