Brain is a typical complex system with characteristics such as self-adaptation, self-organization, and multistability. The activity of the default mode network (DMN), a crucial functional subnetwork of the human brain in resting state, obeys typical non-equilibrium statistical mechanical processes in which the system continually switches among multiple metastable states. Revealing the underlying dynamical mechanism of these processes has important scientific significance and clinical application prospects. In this paper, according to the blood oxygen level dependent (BOLD) signals obtained from functional magnetic resonance imaging (fMRI), we build an energy landscape, disconnectivity graph and transition network to explore the non-equilibrium processes of DMN switching among different attractors in resting state. Taking the activities of high-level visual and auditory cortices for examples, we verify the intimate relationship between the dynamics of DMN and the activity modes of these external brain regions, through comparing the distributions in state space and the algorithms such as XGBoost and deep neural networks. In addition, we analyze the interaction between various DMN regions in the resting state by using the techniques such as compressive-sensing-based partial correlation and convergence cross mapping. The results in this paper may presnt new insights into revealing the dynamics of the intrinsic non-equilibrium processes of brain in resting state, and putting forward clinically significant biomarkers for brain dysfunction from the viewpoint of dynamics.