With　the　rapid　development　of　science　and　technology,　it　has　become　a　common　concern　of　the　society　to　provide　more　intelligent　prosthetics　for　people　with　hand　disabilities　and　help　them　better　restore　their　ability　to　live.　Electromyography,　a　weak　potential　difference　signal　that　can　be　measured　on　the　surface　of　human　skin,　has　gradually　become　the　preferred　signal　source　for　the　control　of　bionic　hand　because　of　its　simple　acquisition　and　convenient　processing.　At　this　stage,　the　further　development　of　EMG　signal　bionic　hand　is　limited　due　to　the　poor　stability,　low　recognition　accuracy,　few　gesture　categories,　large　algorithm　scale　and　other　reasons.　In　view　of　the　above　problems,　through　the　analysis　and　processing　of　EMG　signals,　using　digital　filter,　parallel　multi-scale　CNN　network　and　other　technologies,　this　paper　establishes　a　set　of　EMG　gesture　recognition　algorithm　based　on　parallel　multi-scale　CNN.　The　experimental　results　show　that　the　gesture　recognition　algorithm　proposed　in　this　paper　has　high　accuracy　and　strong　stability.　It　can　classify　more　gesture　categories　and　can　be　applied　to　the　actual　control　of　bionic　hands.　The　algorithm　has　certain　progressiveness.,　which　provides　more　possibilities　for　the　further　development　and　application　of　intelligent　hand　prosthesis.　©　2022　IEEE.