Efficient　and　fair　resource　allocation　is　a　critical　challenge　in　vehicular　networks,　especially　under　high　mobility　and　unknown　channel　state　information　(CSI).　Existing　works　mainly　focus　on　centralized　optimization　with　perfect　CSI　or　decentralized　heuristics　with　partial　CSI,　which　may　not　be　practical　or　effective　in　real-world　scenarios.　In　this　paper,　we　propose　a　novel　hierarchical　deep　reinforcement　learning　(HDRL)　framework　to　address　the　joint　channel　and　power　allocation　problem　in　vehicular　networks　with　high　mobility　and　unknown　CSI.　The　main　contributions　of　this　work　are　twofold.　Firstly,　this　paper　develops　a　multi-agent　reinforcement　learning　architecture　that　integrates　both　centralized　training　with　global　information　and　decentralized　execution　with　local　observations.　The　proposed　architecture　leverages　the　strengths　of　deep　Q-networks　(DQN)　for　discrete　channel　selection　and　deep　deterministic　policy　gradient　(DDPG)　for　continuous　power　control　while　learning　robust　and　adaptive　policies　under　time-varying　channel　conditions.　Secondly,　this　paper　designs　efficient　reward　functions　and　training　algorithms　that　encourage　cooperation　among　vehicles　and　balance　the　trade-off　between　system　throughput　and　individual　fairness.　By　incorporating　Jain＇s　fairness　index　into　the　reward　design　and　adopting　a　hybrid　experience　replay　strategy,　the　proposed　algorithm　achieves　a　good　balance　between　system　efficiency　and　user　equity.　Extensive　simulations　demonstrate　the　superiority　of　the　proposed　HDRL　method　over　state-of-the-art　benchmarks,　including　DQN,　DDPG,　and　fractional　programming,　in　terms　of　both　average　throughput　and　fairness　index　under　various　realistic　settings.　The　proposed　framework　provides　a　promising　solution　for　intelligent　and　efficient　resource　management　in　future　vehicular　networks.