In the optical fiber transmission system, various physical damage effects seriously affect transmission performance. Therefore, it is necessary to monitor the optical performance of the transmission signals to ensure the normal operation of the high-speed optical transmission network. A multi-task optical performance monitoring scheme based on convolutional neural networks (CNNs) is proposed. The intensity profile and intensity fluctuation features are used as the input of the CNN model for the joint monitoring of the modulation format and optical signal-to-noise ratio (OSNR). The results indicate that all the modulation formats (28-GBaud PDM-QPSK/-8QAM/-16QAM/-32QAM/-64QAM) can be accurately identified (identification accuracy is 100%) under OSNR corresponding to the threshold condition of 20% forward error correction (FEC) (bit error rate is 2.4×10^-2). When the intensity profile, the intensity fluctuation, and combination of the two features are used as the model input separately, the mean absolute error of OSNR monitoring is 0.282 dB, 0.245 dB, and 0.165 dB, respectively, and the root mean square error is 0.352 dB, 0.311 dB, and 0.218 dB, respectively. Subsequently, the influence of residual dispersion on the monitoring performance of the proposed scheme is further analyzed.