Power system events such as line trip and fault will destroy the stability of power grid. It is very important for power grid operators to quickly and accurately obtain the category of events, which is helpful to prevent power outages and economic losses. In this paper, the voltage data of phasor measurement units is used to propose a method to quickly detect and recognize the event types in power system. For event detection, by calculating Pearson correlation coefficient, the operation state of power system can be monitored in real time, and events in the system can be found quickly. For event recognition, the measurement data is transformed into two-dimensional images by relative position matrix, and then the event is classified by extracting feature information by convolutional neural network. The proposed method is tested for simulated event cases in the IEEE 39-bus system. Compared with the commonly used time series analysis methods, this method has higher accuracy of event recognition.