The integration of distributed energy and the application of power electronic equipment have promoted the development of new power systems, but also caused complex power quality disturbances (PQDs). In order to accurately identify complex power quality disturbances in modern power systems, a power quality identification method based on Improved Markov Transition Field (IMTF) and Dual-channel attention mechanism convolutional neural networks (DCAMCNN) is proposed. Firstly, the power quality time series signal is transformed into a two-dimensional image through MTF, and the HSV (Hue,Saturation,Value) color space is used to perform secondary color coding on the MTF image. Then, a dual-channel attention mechanism is constructed and fused with a multi-scale convolutional neural network to focus on the important information of PQDs along two channels and suppress lightweight features. Finally, the improved MTF image is input into the constructed model to train and optimize the parameters, and the optimal model is used to output the disturbance classification results. Experimental results show that the proposed method has higher recognition accuracy and anti-noise ability than traditional image transformation methods and other network algorithms.