The　application　of　artificial　intelligence　in　electrocardiogram　(ECG)　diagnosis　holds　substantial　significance.　Most　ECG　classification　methods　concatenate　12-lead　ECG　into　a　2-D　matrix　for　model　input.　This　study　proposed　a　multi-branch　and　multi-class　model　for　arrhythmias　classification.　The　model　utilizes　selective　kernel　block　to　independently　extract　features　from　each　lead,　which　are　fed　into　Bi-LSTM　for　fusion.　Additionally,　batch-free　normalization　module　is　employed　to　reduce　estimation　shift.　Finally,　the　proposed　model　achieved　an　accuracy　of　0.871　and　a　macro　F1　score　of　0.841　in　identifying　nine　types　of　arrhythmias.　©　2024　IEEE.