To address the problem that traditional methods cannot accurately identify household loads containing high-order harmonics, a non-intrusive load identification method based on the fusion of multiple features containing V-I trajectory matrix, power and high-order harmonics is proposed. Firstly, the V-I trajectory matrix, power characteristics and harmonic characteristics of 11 kinds of family load are analyzed. Secondly a hybrid feature matrix construction method based on pixel image conversion is proposed. The power and high-order harmonic characteristics of the load are combined with the basic V-I pixel trajectory through binary coding conversion,which enriches the characteristic information of samples. Thirdly,the mixed feature matrix is used as the input of the convolutional neural network to realize the accurate recognition of the household load identification. In the calculation example,the algorithm proposed in this paper can accurately distinguish two loads of heaters and hair dryers with similar power characteristics but different high-order harmonic content. It achieves an identification accuracy rate of more than 93% for all types of household loads. This algorithm provides the technical support for accurately investigating potential safety risk of household electrical loads containing high-order harmonics in engineering application.