I’ve understood and accepted @the_fox’s answer. I want to write something for myself.
I was stuck exactly because I didn’t know the idea of non-generator and the fact that the Frattini subgroup of a group is the set of all non-generators of this group. I searched on this site, but nearly all of the proofs used Zorn’s Lemma. I want to add a proof in the finite case without using the lemma.
Define $X$ to be the set of all non-generators of $G$. We want to prove $X=\Phi(G)$.
Since $1$ is a non-generator of $G$, $X$ is not empty.
For any $x\in X$ and any maximal subgroup $M$ of $G$, by the definition of non-generator, $\langle M,x\rangle=M$; otherwise, if $M<\langle M, x\rangle$, then $\langle M,x\rangle=G$ (since $M$ is maximal), contradicting the definition of non-generator. Hence $x\in M$ and it implies $X\subseteq \Phi(G)$.
For any $g\in \Phi(G)$ and any maximal subgroup $M$ of $G$, since $g\in M$, we have $\langle M,g\rangle=\langle M\rangle=M\neq G$. For any subgroup $H$ of $G$, $H$ is contained in at least one of the maximal subgroups of $G$. Without generality, $H\le M$. Then $\langle H,g\rangle\le\langle M,g\rangle\neq G$. Hence $g$ is a non-generator and it implies $\Phi(G)\subseteq X$.
Thus $X=\Phi(G)$.