Abstract: In power system, password is one of the important ways of authentication. The traditional weak password scanning scheme mainly focuses on the password length, the same letter combination, the correlation between password and personal information, but does not pay attention to the weak password with morphological characteristics based on keyboard coordinates. In this paper, the password was transformed into 28×28 image according to the position of the keyboard, and the morphological features of the password were learned by convolution neural network, so as to effectively identify the weak password with morphological characteristics. Compared with the existing password strength evaluation methods based on N-gram Markov model method and Kaspersky tester, this scheme has higher accuracy and significant recognition precision, which can guarantee the password security of power system.