Like albatross, unmanned aerial vehicles can significantly make use of wind gradient to extract energy by the flight technique named dynamic soaring. The research aims to develop a general optimization method to compute all the possible patterns of dynamic soaring with a small unmanned aerial vehicle. A direct collocation approach based on the Runge-Kutta integrator is proposed to solve the trajectory optimization problem for dynamic soaring. The optimal dynamic soaring trajectories are classified into two patterns: closed trajectory pattern and travelling trajectory pattern by applying terminal constraints of zero horizontal displacement and a certain travelling direction, respectively. Using different terminal constrains for heading angle and initial guesses in the optimization process, the former pattern can be divided into two subtypes: O-shaped and 8-shaped trajectories, while the latter one is divided into C-shaped, α-shaped, S-shaped and -shaped trajectories. The characteristics of these patterns and the correlation among patterns are analyzed and discussed.