As a new hyperspectral remote sensing technology, the thermal infrared hyperspectral remote sensing has the ability to observe the surface day and night. Instead of the surface temperature retrieval, the ultimate goal of the hyperspectral thermal infrared remote sensing data is to retrieve surface emissivity which can be used for mineral detection, ground parameters inversion and target identification and so on. Due to the inevitable atmospheric influence during the data acquisition process of thermal infrared hyperspectral imager, the atmospheric radiation information is included in the hyperspectral thermal infrared remote sensing data. The atmospheric correction with high precision is the premise of effectively implementing the subsequent applications. The aircraft, especially the unmanned aerial vehicle rising in recent years, is an important platform for remote sensing because of its flexibility, immunity to the cloud cover and high spatial resolution. The aircraft flies through the atmosphere when working, which makes the direct measurement possible for the atmosphere. On the basis of summarizing the features of the atmospheric correction methods for the hyperspectral thermal infrared remote sensing data, and the principle and process of atmospheric radiative transfer, the paper proposed an overall vision for the atmospheric correction based on the measured atmospheric downwelling radiance at the aircraft flight height, designed an atmospheric correction system for airborne thermal infrared imaging spectrometer, and stated the atmospheric correction procedure of hyperspectral thermal infrared remote sensing data using the system. The proposed system and procedure for atmospheric correction are able to acquire the atmospheric downwelling thermal infrared radiation information in real time, so without too many assumptions, it is possible to dramatically increase the precision of the atmospheric correction for the airborne hyperspectral thermal infrared remote sensing data and gives a promise to the objectivity of the data.