In　this　paper,　reinforcement　learning　(RL)　based　cognitive　anti-jamming　system　employing　full　duplex　tactical　radio　is　investigated　under　electromagnetic　spectrum　warfare　scenario.　Firstly,　the　analytical　expressions　of　jamming　sensing　based　on　improved　energy　detection　are　derived　to　calculate　the　reward　metric.　Then,　we　propose　the　multidomain　anti-jamming　strategies　based　on　different　learning　algorithm　and　the　accurate　reward.　Simulation　results　indicate　that　learning-based　cognitive　anti-jamming　strategies　may　increase　about　25%　of　the　throughput　of　tactical　radio.　Moreover,　the　upper　confidence　bound　and　Thompson　sampling　strategies　almost　have　the　same　performance　and　they　are　superior　to　other　RL　anti-jamming　schemes.