In cognitive radio networks, allocation of the sensing duration is of critical importance for efficient spectrum sensing. Conventional optimization of sensing duration is limited in two primary channels as well as same constant thresholds for all the primary channels. In this paper, we consider multiple primary channels and formulate the problem of sensing duration allocation as a convex optimization with respect to both the sensing durations and detection thresholds. Then, based on Penalty method and Newton's method, a searching algorithm is proposed to obtain the optimal solution. We show via simulation results that our scheme substantially improves the transmission opportunity of cognitive user compared with existing ones.