Feature extraction method and device based on stacked auto-encoder and terminal equipment

Abstract

The invention provides a feature extraction method and device based on stacked auto-encoders and terminal equipment, the method is applied to the technical field of feature extraction, and the methodcomprises the following steps: setting the number k of current auto-encoders of the stacked auto-encoders to be equal to 1; determining structural parameters of a k-auto-encoder based on an improved shuffled frog leaping algorithm; determining a reconstruction error of the k-auto-encoder based on the structural parameters of the k-auto-encoder; if the reconstruction error of the k-auto-encoder isgreater than a preset threshold, adding an auto-encoder in the stacked auto-encoder, enabling k to be equal to k + 1, and returning to execute the step of determining the structural parameters of thek-auto-encoder based on the improved shuffled frog leaping algorithm; and if the reconstruction error of the k-auto-encoder is not greater than a preset threshold, determining that the stacked auto-encoder is trained, and performing feature extraction on the data based on the trained stacked auto-encoder. According to the feature extraction method and device based on the stacked auto-encoder and the terminal equipment provided by the invention, the speed and precision of feature extraction can be improved.

Description

technical field [0001] The present invention belongs to the technical field of feature extraction, and more specifically relates to a feature extraction method, device and terminal equipment based on stacked autoencoders. Background technique [0002] Stacked autoencoders, as a typical architecture of deep learning, can reduce the dimension to obtain a series of simple high-order features that can well express the input data through layer-by-layer greedy learning, and have obvious advantages in feature extraction of data. Advantage. [0003] In the existing feature extraction process, when the stacked autoencoder is trained, the number of stacked autoencoders in the stacked autoencoder is difficult to determine, and the structural parameters of each autoencoder are also randomly initialized, which leads to In the training of stacked autoencoders, the convergence speed of stacked autoencoders is slow and the convergence accuracy is low. Among them, the slow convergence spee...

AI Technical Summary

Problems solved by technology

[0003] In the existing feature extraction process, when the stacked autoencoder is trained, the number of stacked autoencoders in the stacked autoencoder is difficult to determine, and the structural parameters of each autoencoder are also randomly initialized, which leads to When training the stacked autoencoder, the stacked autoencoder has slow convergence speed and low convergence accuracy

Among them, the slow convergence speed of the stacked autoencoder will lead to a large network depth value of the trained stacked autoencoder, which will affect the speed of feature extraction, and the low convergence accuracy of the stacked autoencoder will affect the accuracy of feature extraction.

Images