Abstract: The rapid screening and identification of mild depression have important practical significance. The thesis uses resting state EEG signals to explore effective methods of machine recognition for mild depression, aiming to find out the EEG characteristics and classification algorithms that are closely related to depression. First, add time window segments to the original EEG data of each link to calculate the EEG activity, mobility, and complexity of both depression patients and normal people; then use Burg algorithm and wavelet transform separately to extract frequency domain features and time-frequency non-linear features from each EEG signal; finally, the support vector machine（SVM） algorithm was used to classify depression EEG, and the effects of different time Windows, lead combinations, feature combinations, rhythm combinations and machine learning algorithms on the recognition results were analyzed. The experimental results show that the depression can be identified with accuracy rate of 94.24%, recall rate of 92.35%, and precision rate of 96.23%, by using a support vector machine classifier, selecting a 20 s time window, the combination of O2 and T5 leads, as well as the combination of activity, mobility, wavelet energy entropy, wavelet singular entropy feature and EEG alpha, beta, gamma rhythm combination.