To　meet　the　spectrum　resource,　the　spectrum　sensing　technology　in　the　Cognitive　Radio　(CR)　is　developed　rapidly.　The　multitude　of　diverse　challenges　are　posed　by　most　existing　spectrum　sensing　methods,　due　to　the　demand　of　the　prior　knowledge　of　signal　and　the　high　computational　complexity.　A　spectrum　sensing　method　based　on　Attention-Based　CNN　is　proposed　in　this　paper.　The　spectrum　sensing　is　treated　as　a　binary　classification　problem.　The　derived　models　are　validated　by　extensive　experiments　with　respective　results　from　computer　simulations.　It　is　demonstrated　that　the　proposed　method　can　achieve　high　recognition　accuracy　and　recognition　speed　under　the　condition　of　low　Signal　to　Noise　Ratio　(SNR).　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.