In the traditional non-invasive load monitoring (NILM) algorithms, the identification accuracy is enhanced with the increased network scale while sacrificing the calculation speed, which restricts the efficiency of the load identification. In this study, a multi-feature (active/reactive power and current peak-to-peak value) fusion algorithm is proposed, which can achieve enhanced identification accuracy with a smaller network scale while maintaining the calculation speed. The features of the power and current amplitudes of the loads are transformed into the values of red-green-blue (RGB) color channels by color coding and then fused into the V-I trajectory features. After that, the true-color feature image with higher discrimination is generated and input into the convolutional neural network (CNN). The testing results on the PLAID data set indicate that in comparison with the traditional load identification algorithm, the algorithm proposed in this study performs higher identification accuracy with a smaller neural network parameter scale, which significantly improves the identification efficiency.