Abstract: The three-dimensional combustion field in the large-scale furnace is closely related to the safety, economy, and pollutant emission level of combustion process. Compared with other visualization monitoring technology of three-dimensional temperature field, such as acoustic tomography and absorption spectrum tomography, the thermal radiation imaging method has a compact system and easy to implement, with high temporal and spatial resolution, and has great application potential. In this paper, the principles of thermal radiation imaging in large-scale furnaces were introduced, the ill-posedness of the inverse problem of radiative transfer was analyzed, and the progress of the algorithm for the inverse problem of radiative transfer was reviewed. The construction of a large-scale furnace thermal radiation imaging model uses a thermal radiation imaging matrix to link the radiation energy distribution received by the sensor with the temperature field and radiation characteristics of the medium and wall. The critical step to calculating the thermal radiation imaging matrix is to obtain the scattering or reflection share of the medium elements and wall elements. The DRESOR method and the inverse Monte Carlo method can solve the problem. Judging by the condition number of the thermal radiation imaging matrix, the inverse problem of radiative transfer is ill-posedness. As a result, the solution is non-uniqueness or even non-existence, and minor measurement errors will cause the instability of the reconstructed temperature field. Solving this kind of ill-posed problem is mainly divided into optimization methods and regularization methods. Optimization methods can divide into traditional optimization methods and intelligent optimization methods. Traditional optimization methods are based on gradient calculations, which reduce the objective function through repeated iterative calculations, such as the Least-Squares and Conjugate Gradient. However, this type of method relies heavily on the initial value, requires the derivative of the objective function, and cannot obtain the optimal global solution. Intelligent optimization methods are based on probabilistic search, which starts from a certain random solution and looks for the optimal solution in the solution space with a certain probability according to the corresponding algorithm mechanism. There is no need to know the exact mathematical model of the optimization problem and not necessarily solve the gradient of the objective function. It can also divide into bio-colony simulation and bio-individual simulation. The former includes Particle Swarm Optimization(PSO),Genetic Algorithm(GA),and Ant Colony Algorithm(ACO),etc. It needs to construct an objective function and takes a long time to search for the optimal solution. The latter includes Artificial Neural Networks(ANN),Support Vector Machines(SVM),etc. The mathematical equations of the mapping relationship of the prediction problem are not necessary to know in advance; the result closest to the actual output value can obtain after training. But the quality of the training data set is one of the critical factors affecting the prediction accuracy. Regularization methods are also common for ill-posed problems, including Tikhonov regularization(TR),Truncated Singular Value Decomposition(TSVD),etc. The solution of a family of well-posed problems adjacent to the original ill-posed problem is used to approximate the solution. This method has been used to reconstruct the three-dimensional temperature field in coal-fired furnaces and has high reconstruction accuracy and efficiency. Although thermal radiation imaging needs to consider the influence factors such as optical thickness, it can better reproduce the distribution characteristics of the actual temperature field in the furnace within a certain scope of application. And it considers the radiation transfer equation in the three-dimensional space, which is essentially a three-dimensional monitoring technology. The development of imaging technology(light field camera, multi/hyperspectral imager, etc.) has also pointed out a new development direction for thermal radiation imaging to reconstruct the three-dimensional temperature field in large-scale furnaces.