Each　fog　node　interacts　with　data　from　multiple　end-users　in　mobile　fog　computing　(MFC)　networks.　Malicious　users　can　use　a　variety　of　programmable　wireless　devices　to　launch　different　modes　of　smart　attacks　such　as　impersonation　attack,　jamming　attack,　and　eavesdropping　attack　between　fog　servers　and　legitimate　users.　The　existing　research　in　MFC　lacks　in　the　contributions　of　defense　of　smart　attack　and　also　requires　in　the　discussions　of　subjective　decision　making　by　participants.　Therefore,　we　propose　a　smart　attack　defense　scheme　for　authorized　users　in　MFC　in　this　paper.　First,　we　construct　a　static　zero-sum　game　model　between　smart　attackers　and　legitimate　users　based　on　prospect　theory.　Second,　the　double　Q-learning　(DQL)　is　proposed　to　restrain　the　attack　motive　of　smart　attackers　in　the　dynamic　environment.　The　proposed　DQL　method　generates　the　optimum　defense　choice　of　legitimate　users　against　smart　attacks　so　that　they　can　efficiently　determine　whether　to　use　only　physical　layer　security　(PLS)　to　avoid　those　smart　attacks.　We　use　our　scheme　to　contrast　with　the　basic　schemes,　i.e.,　Q-learning　scheme,　the　Sarsa　scheme,　and　the　greedy　strategy.　Experiment　results　prove　that　the　proposed　scheme　can　enhance　the　utility　of　legitimate　users,　restrain　the　attack　motive　of　smart　attackers,　and　further　provide　better　security　protection　in　the　MFC　environment.