The invention discloses a CNN-based transformer fault diagnosis method. The method comprises the following steps of 1, obtaining a monitoring image of a transformer; 2, inputting the monitoring imageinto a CNN judgment model, and obtaining a probability value of the monitoring image being a transformer in a normal state; And step 3, judging whether the probability value is lower than a set probability threshold value or not, and if yes, judging that the transformer is in a fault state. According to the invention, the deep neural network is used to determine and diagnose the fault state of themonitoring image of the transformer; According to the method, the calculation complexity of traditional fault diagnosis is reduced, the efficiency of transformer fault identification and diagnosis inthe transformer substation and the accuracy of a diagnosis result can be effectively improved, the robustness is high, and monitoring images of different transformer substation backgrounds can be processed.