To enhance reconstruction performance in fluorescence molecular tomography, a joint-norm and a Laplacian manifold regularization model that combined both sparsity and spatial aggregation information was utilized for light source reconstruction. In this report, sparse reconstruction by separable approximation (SpaRSA) was developed to investigate the joint model (SpaRSA-resolved Laplacian manifold regularization model, SpaRSALM). To improve the convergence speed of the SpaRSALM algorithm, a warm-start strategy was applied for light source reconstruction. The experimental results show that the SpaRSALM algorithm solved the joint model problem and improved the contrast to noise ratio (CNR) from 6.45 to 9.18 compared to using the SpaRSA algorithm to solve for the -norm regularization model. In addition, the reconstruction of the SpaRSALM algorithm using the warm-start strategy (compared to without the warm-start strategy) required 50.10 s (as opposed to 101.84 s). The accuracy and speed of light source reconstruction were significantly improved, and better reconstruction results were achieved using the presented method.