Mental　arithmetic　is　a　basic　cognitive　function　of　human　brain,　mental　arithmetic　is　an　important　cognitive　function　of　brain,　which　is　also　considered　as　the　core　of　human　logical　thinking.　fMRI　provides　convenience　for　mental　arithmetic　cognitive　function　research　because　of　its　non-invasiveness　and　convenience,　More　and　more　experiments　are　devoted　to　the　clear　understanding　of　mental　arithmetic,　and　the　classification　of　cognitive　tasks　will　contribute　to　a　more　comprehensive　understanding　of　the　behavior　of　organisms　and　the　decoding　of　neural　signals.　We　recruited　21　subjects　and　took　block　design　of　fMRI　in　the　experiment,　In　this　paper,　seeds　are　extracted　and　characterized　by　partial　correlation　connection,　build　functional　brain　network　(FBN).　At　present,　the　deep　learning　technology　represented　by　CNN　is　increasingly　applied　in　the　analysis　and　classification　of　neural　image　data,　CNN　was　used　to　classify　the　brain　functional　connectivity　network,　at　the　same　time,　with　several　other　machine　learning　methods　classifying　mental　arithmetic　comparison　effects　and　analyzed.　The　classification　results　show　that　in　the　data-driven　classification,　CNN-based　method　have　the　best　classification　effect,　reaching　98%,　which　is　more　obvious　than　the　traditional　machine　learning　method.　©　2019,　Springer　Nature　Switzerland　AG.