Depression　has　become　a　high　incidence　in　today＇s　society　due　to　its　complex　pathogenesis　and　difficult　to　cure,　and　there　is　currently　no　recognized　auxiliary　means　that　can　be　used　to　diagnose　depression.　In　this　paper,　the　EEG　signals　obtained　by　brain-computer　interface　technology　are　analyzed　and　studied,　and　the　characteristic　matrix　is　constructed　by　extracting　the　log-Mel　energy　of　the　16-electrode　channels　of　EEG　by　taking　advantage　of　the　significant　difference　in　brain　region　activity　between　depressed　patients　and　non-depressed　patients.　The　dataset　we　used　in　this　paper　contains　16　confirmed　depressed　patients　and　16　healthy　participants.　The　four-layer　convolutional　neural　network　model　constructed　in　this　paper　can　well　achieve　depression　recognition,　and　the　classification　accuracy　of　the　model　can　reach　98.　　©　2023　IEEE.