Fluorescence molecular tomography has become a promising technique for in vivo small animal imaging and has many potential applications. Due to the ill-posed and the ill-conditioned nature of the problem, Tikhonov regularization is generally adopted to stabilize the solution. However, the result is usually over-smoothed. In this letter, the third-order simplified spherical harmonics approximation to radiative transfer equation is utilized to model the photon propagation within biological tissues. Considering the sparsity of the fluorescent sources, we replace Tikhonov method with an iteratively reweighted scheme. By dynamically updating the weight matrix, L1-norm regularization can be approximated, which can promote the sparsity of the solution. Simulation study shows that this method can preserve the sparsity of the fluorescent sources within heterogeneous medium, even with very limited measurement data.