A swarm initialization method is proposed for modified shuffled frog leaping algorithm (SFLA) for linear combination cooperative spectrum sensing in cognitive radio. The solution obtained by modified deflection coefficient optimization is included in the initial swarm of SFLA, thus improving the performance of the algorithm at the early stage of the search. Simulations show that compared with the traditional swarm initialization technique, the proposed swarm initialization can obtain expected solutions faster, which means that the proposed technique can save computation time and is more suitable for real time applications.