Classification model training and object classification method and device

Abstract

The embodiment of the invention provides a classification model training and object classification method and device. In the training method, a sample with a classification label is obtained; in the embedding layer, a feature vector of a sample and a tag vector of a classification tag are determined. And in the convolution layer, based on a plurality of convolution windows with different widths, multiple convolution processing is carried out on the feature vector of the sample to obtain a plurality of convolution results. And in the pooling layer, the similarity between each convolution resultand the label vector of the classification label is calculated, and an attention weight value corresponding to each convolution result is determined based on the calculated similarity. And based on the attention weight value corresponding to each convolution result, weighted average pooling operation is performed on each convolution result to obtain a pooling result. And the pooling result is taken as a sample representation vector of the sample, and prediction loss is determined at least based on the sample representation vector and the label vector of the classification label. And parameters of the classification model are adjusted based on the prediction loss.

Description

technical field [0001] One or more embodiments of this specification relate to the field of computer technology, and in particular, to a classification model training, object classification method and device. Background technique [0002] Object classification refers to predicting which category the object to be classified belongs to in each specific category under a specific classification system through a pre-trained classification model. The classification model here may be, for example, a convolutional neural network, and the convolutional neural network may include a convolutional layer and a pooling layer. [0003] Taking the classification model as a convolutional neural network as an example, in the traditional model training method, in the convolution layer, local features are extracted from samples through convolution operations. In the pooling layer, global features are extracted from local features by max pooling or average pooling operations. However, since ma...

AI Technical Summary

Problems solved by technology

However, since the maximum pooling will strengthen the local features of the sample, this part of the features may not be important information related to object classification

Furthermore, average pooling dilutes the effective features

Therefore, traditional training methods often cannot extract effective global features from samples, which will affect the accuracy of the trained classification model.

Images