An Infrared Temperature Measurement Method Based on Improved Convolutional Neural Network under Water Mist Interference

Infrared Thermography (IRT) has been widely used for temperature measurement of high-temperature objects due to its characteristics of non-contact and broad temperature measurement range. However, interference factors such as water mist can cause deviations in the process of temperature measurements, and the impact becomes increasingly evident as the temperature increases. In this study, an infrared temperature measurement method based on improved convolutional neural network (CNN) is proposed. Improved CNN model is established by combining ResNet34 to extract the multi-channel features, incorporating a Squeeze-and-Excitation module to adjust the weights of each feature channel adaptively. The weighted features are concatenated with the external features of the image to regress the water mist transmittance. Then, the temperature measurement result is compensated by combining the water mist particle cluster mechanism model and transmittance. Experimental results illustrate that the proposed method is capable of improving the accuracy of infrared temperature measurement for high temperatures.