The theory of compressed sensing is integrated with deflection angle revision iteration technique for deflection tomography reconstruction from few view- angles projections. A new iterative reconstruction algorithm with deflection angle compressed sensing revision is developed to improve the quality of reconstruction distribution from incomplete projection data. The algorithm, which combines the sparsity of the object field gradient l1 norm, is adjusted by total variation with the steepest descent method. A compressed sensing weighting factor is employed in every iteration step to optimize the reconstruction. In numerical simulation, the new approach produces reliable reconstruction for asymmetrical temperature distribution from 6 view- angles projections. The error analysis provides a good indication of the precision and convergence of the algorithm. Compared with our previous algorithm for deflection tomography, the new algorithm turns out to be more accurate.The maximum value error is reduced by 18.32% and the normalized distance error is reduced by 46.67% when the new reconstruction algorithm is employed. The efficacy of the new algorithm has been assessed by reconstructing a flame temperature field.