Abstract: In order to solve the problem of increasing computation and damaging image structure caused by principal component analysis (PCA) in convolution kernel simulation, a 2DPCANet face recognition method based on improved perceptual hashing algorithm is proposed. Based on the principal component analysis network, the method replaces the calculation process of principal component analysis in the original network architecture with two-dimensional principal component analysis, extracts the local features of the image with binary hash and block histogram, and finally classifies the face image with support vector machine and limit learning machine. Based on FERET data set and LFW face database, the recognition effect of the proposed method and several comparison methods in face recognition task is compared. The experimental results show that the proposed method improves the recognition rate by 10%~21% compared with other methods.