Abstract: Aiming at the problem that when the local linear embedding algorithm used Euclidean distance to calculate the similarity of unaligned samples, it was greatly affected by data position difference, resulting in low measurement accuracy and affecting the accuracy of the algorithm feature extraction, we proposed a local linear embedding algorithm based on information entropy measurement. Firstly, the degree of confusion among sample features was counted by information entropy to improve the accuracy of dividing local neighborhoods. Secondly, a local reconstruction model was established to dig out the essential structure of the manifold. Finally, the local structure was used to build a low-dimensional reconstruction model to obtain the salientfeatures of the samples. The experiment on the bearing data set proves the effectiveness of the algorithm in feature extraction.

Key words: local linear embedding, feature extraction, information entropy, data alignment