Abstract: Accurately locating forced oscillation disturbance sources is the key to eliminate forced oscillation and restore normal operation of power system. A location method of forced oscillation disturbance source based on smoothed pseudo Wigner-Ville distribution（SPWVD） image and deep transfer learning is proposed. Firstly, for the forced oscillation signals, SPWVD is adopted to represent the forced oscillation characteristic information of the whole power system in the form of image. Then, through deep transfer learning, the image recognition knowledge from other fields is transferred into the power system to mine the relationship between SPWVD images and the location of disturbance sources, which guarantees the training accuracy and improves the training efficiency. Finally, the effectiveness of the proposed method is verified in WECC 179-bus system, and it has higher accuracy than traditional machine learning method. In addition, considering the noise, the start time of recording and data length in the oscillation data, the accuracy and noise resistance of the proposed method are verified under the condition of forced oscillation induced by load fluctuation and changing system topology.