Abstract: With the rapid development of the smart grid, two-way interactive intelligent power utilization has gradually become one of the main development goals of power systems. The non-intrusive load monitoring (NILM) technology is crucial in obtaining the user-side load information for intelligent power utilization. Based on the appliance switching on/off event detection, this paper proposes a non-intrusive load identification method harmonizing with load high-frequency features. Firstly, it extracts load high-frequency current and voltage waveforms, plots the U-I characteristic curve as a gray-scale diagram, and converts the cycle current and the cycle instantaneous active power into characteristic gray-scale diagrams by Gramian Angular Field and Markov Transition Field algorithms. Secondly, it puts the three grayscale images into the R-channel, G-channel, and B-channel of the picture to form an RGB picture. Finally, it trains the feature RGB pictures using the ShuffleNetV2 neural network. The experiment verifies that the method can identify household appliances efficiently with a wide variety of appliances with fast convergence and high accuracy.