End-to-end zero sample learning method based on semantic description

Abstract

The invention discloses an end-to-end zero sample learning method based on semantic description, and the method comprises the steps: automatically constructing category semantic description characteristics, and completing the generation of category description semantic vectors through employing a long-short-term memory neural network. Through combined training of image feature extraction and zerosample migration modeling, a technical scheme of an end-to-end (combined training) zero sample learning model based on semantic description is realized, and the problems of non-visual property and ambiguity caused by using a single word vector are solved. And the image feature extraction module and the zero sample migration modeling module are combined for training, so that the end-to-end model ismore convenient and quicker. Semantic vector construction can be customized for different scenes, and the method is more accurate and efficient.

Description

technical field [0001] The invention relates to an end-to-end zero-sample learning method based on semantic description, which belongs to the field of computer technology. Background technique [0002] For a long time, computer vision (Computer vision), natural language processing (Natural language processing), speech recognition (Speech recognition), etc. have mainly focused on solving supervised learning and semi-supervised learning problems, and supervised learning methods in the past period of time Considerable breakthroughs have been made, such as face recognition, vehicle detection, license plate recognition, etc., have even been put into actual production and life. Efficient and accurate classifiers have brought great convenience to people's daily life and greatly reduced the cost. field costs. As far as the field of computer vision is concerned, with the continuous innovation of deep convolutional neural networks and the substantial improvement of computer processin...

AI Technical Summary

Problems solved by technology

At the same time, the feature of the picture is too dependent on the data set, and the separation training leads to zero-sample feature learning and loss alignment network can not reach the best state, and the feature extraction network has over-fitting phenomenon, modeling zero-sample migration caused great difficulty

Images