The tunable diode laser absorption spectroscopy tomography (TDLAST) is an important non-intrusive combustion monitoring technology. However, the ill-posedness of the inverse problem of TDLAST causes large errors in gas absorbance densities reconstructed by using traditional algorithms. In this article, the expected patch Log likelihood is introduced into TDLAST to solve the inverse problem. The Gaussian mixture model (GMM) regularized temperature reconstruction technique (GMMTRT) is proposed to image temperature distribution in the combustion field. This algorithm models the distribution characteristics of local absorbance density with GMM, and solves the inverse problem of TDLAST with GMM regularization via the half quadratic splitting method. Both the simulation with data generated from the fire dynamics simulator and the real experiment with the lab-scale TDLAST system show that GMMTRT can clearly indicate the temperature profile in the region of interest with correctly located flame peaks. Compared to the temperature reconstruction algorithm with Tikhonov regularization and the simultaneous algebraic reconstruction technique, GMMTRT can reduce the reconstruction error by15. 42% ~ 36. 16% and 23. 10% ~ 44. 79% , respectively.