The　encephalography　(EEG)　based　epilepsy　detection　can　be　automatically　done　by　deep　learning　approach　and　has　gained　more　and　more　attention,　while　it　is　a　challenging　issue　to　seek　and　use　the　most　critical　channels.　In　this　paper,　we　propose　a　novel　seizure　detection　method　based　on　channel　selection　and　a　dynamic　convolutional　neural　network　(CNN).　All　channels　of　EEG　signals　are　firstly　preprocessed　by　filtering　and　segmentation.　Then,　the　high-frequency　oscillation　(HFOs)　characteristics　of　the　EEG　during　epileptic　seizures　are　statistically　calculated　to　initially　locate　the　onset　region　of　epileptic　seizures　and　determine　its　central　channel,　whose　dynamic　correlation　indexes　(DCRI)　are　obtained　by　using　the　mutual　information　and　Gini　index　between　the　center　channel　and　any　other　one,　and　they　are　ranked　in　a　descending　order.　Furthermore,　a　dynamic　CNN　with　channel　attention　mechanism　is　designed　and　combined　with　the　dynamic　correlation　index　of　partial　channels　to　realize　epilepsy　detection.　The　experiments　are　conducted　on　a　public　dataset　CHB-MIT,　the　average　accuracy　of　five　subjects　is　98.89%,　which　was　also　examined　by　Sensitivity.　The　results　show　that　the　proposed　method　has　the　ability　to　dynamically　change　the　parameters　of　dynamic　CNN　to　match　the　characteristics　of　input　channels,　which　enhances　the　adaptability　of　channel　selection　and　improve　the　epilepsy　detection　effect　as　well.　©　2023　Technical　Committee　on　Control　Theory,　Chinese　Association　of　Automation.