<p>Complex background noise poses significant challenges to infrared thermographic temperature field monitoring in high-speed aircraft. This study proposes a multi-source fusion denoising method integrating infrared thermography, simulated experiment, and sensor temperature. The proposed method employs Gaussian filtering to extract frequency domain characteristics from both simulated experimental temperature field image and infrared thermal temperature field image. Sensor data is utilized as the optimization target for self-adaptive threshold selection. A frequency domain fusion method is subsequently applied to reconstruct the structural temperature field. Leveraging thermal simulation data from typical structure, a multi-source dataset comprising infrared thermography, simulated experiment, and sensor temperature was constructed. Denoising modeling is conducted on infrared thermal temperature fields under diverse noise types and distributions. The analytical result demonstrates that, compared to traditional filter noise reduction method, the proposed method exhibits superior noise reduction capabilities across diverse noise types, significantly enhancing both the signal-to-noise (SNR) ratio of temperature field images and the measurement accuracy of reconstructed temperature distributions. For temperature fields contaminated by Gaussian noise, the mean absolute error (MAE) is reduced from 39 °C to 21 °C after traditional method, and further decreased to 0.44 °C with the proposed method; under higher salt and pepper noise distribution levels which MAE is 17 °C, the MAE further decreases to 0.42 °C.</p>