With the rapid increase in the number of deep packet inspection rules, it is necessary to store deterministic finite automata (DFA) representations of regular expressions efficiently in order to meet the practical requirements of network processing. First, a new hybrid FSM construction method is proposed for compressing the states of DFA. DFAs are built in different ways for the regular expressions. By analyzing the states of the converted DFAs, the distinguished complexities of DFAs become noticeable. This leads to a change in state of the DFA from a quadratic/exponential expression to a linear expression. Next, an efficient compressing algorithm, called Weighted Delayed Input DFA (WD²FA), is proposed for state transitions of the DFAs. This algorithm can reach a reduction rate of about 95% for the regular expressions with any complexity. The analysis shows that the performance of the WD²FA is better than the delayed input DFA (D²FA), and D²FA is a special case of WD²FA with weight 0. The experimental results show that the number of states for the FSM can be controlled at the level of linearity, and transitions are reduced to 7% based on the compression states.