Neural network model training method based on pyramid pooling and long-term memory structure

Abstract

The invention provides a neural network model training method based on a pyramid pooling and long-term memory structure. The neural network model training method comprises the following steps: a, establishing a neural network model based on the pyramid pooling and long-term memory structure; b, pre-training the model; c, carrying out adaptive training on the model; according to the method providedby the invention, a new neural network structure model is provided, the model can extract and store feature information of each sample in a sample image in the source domain, and pre-training and adaptive training are provided for the model; meanwhile, a common task knowledge vector t * is introduced between the source domain encoder Ms and the target domain encoder Mt to serve as an auxiliary variable, and the auxiliary variable is used for storing feature information of the source domain sample images and used for calibrating feature output of all the sample images; and then calibrating thefeature information of the sample images in the target field in the adaptive training step, thereby improving the similarity of image distribution in different fields.

Description

technical field [0001] The invention relates to the technical field of image classification, in particular to a neural network model training method based on pyramid pooling and long-term memory structure. Background technique [0002] Among methods for automatically classifying images by a computer, a method commonly used today is a machine learning method. Humans provide sample t image data, and then a training algorithm trains a model to classify images. [0003] The specific training process of the machine learning model is as follows: first, manually collect sample images, and classify the sample images, mark each image into the correct category, and then divide all sample images into a training set and a test set. The training algorithm uses the training set to train the model in order to achieve the best classification accuracy in the sample images that the model has not seen before. The effect of the training algorithm, that is, the classification accuracy of the m...

AI Technical Summary

Problems solved by technology

[0009] The purpose of the present invention is to provide a neural network model training method based on pyramid pooling and long-term memory structure to solve the problem of low classification accuracy in the prior art due to the different distribution of training set and test set data

Images