Major depression is characterized by low mood and slow thinking, and emotion is the attitude experience of object and the corresponding behavioral response of human brain neurons. Many articles have designed classification network of depression and emotion classification, but the network function is single, can only complete a single classification task, and does not combine mental illness with human emotions, language expression, blinking and other behaviors well. The article explores the correlation of index characteristics between emotion classification and depression diagnosis, then designs a network to verify the feasibility of diagnosing depression through emotion across categories and datasets. The differential entropy is extracted as the input feature of the network, and the convolutional neural network is used to study the emotion classification of SEED-IV and the MODMA emotion proportion. Analyze the microstate parameters of the two datasets, samples with the same microstate type are analyzed and the correlation between the two microstate is explored. The difference of α and γ rhythm classification results and microstate correlation coefficients can be used to classify emotions and diagnose depression. After verifying that parameters in both α and γ rhythms show the correlation between emotion and depression, the design experiment proves that abnormal brain characteristics of patients with depression can be captured by adding microstate features, and the diagnosis of depression can be completed by adding correlated microstate parameters to CNN used for emotion recognition.