An image classification method based on Echo State Network

Abstract

As a traditional neural network architecture, Echo State Network(ESN) has achieved good results in time series forecasting. However, in the past ten years, few people have conducted research on its application to images, especially in classification tasks. Echo State Network is a kind of Recurrent Neural Network, it has the advantages of simple structure and convenient training. It only needs to train its output weight <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$w_{out}$</tex> , which saves a lot of time and resources. In this article, we input the whole picture into ESN in multi-input mode to classify images, and use the traditional Long Short-Term Memory(LSTM) model to compare experiments with it, and the results found that esn has better stability and accuracy.