DEEP METRIC LEARNING METHOD BASED ON HIERARCHICAL TRIPLET LOSS FUNCTION, AND APPARATUS THEREOF

摘要

Provided in the present application are a deep metric learning method based on hierarchical triplet loss functions, and an apparatus thereof, the method comprising: constructing a hierarchical category tree based on triplet loss functions; hierarchising the triplet loss functions to obtain hierarchical triplet loss functions; using the hierarchical triplet loss functions to train a deep neural network; acquiring target image extract features and performing an image search to obtain a target search image. The present application, by means of pre-constructing a hierarchical category tree and obtaining hierarchical triplet loss functions on the basis of the hierarchical category tree, then training a neural network by means of the hierarchical triplet loss functions, features having already been extracted, and performing an image search, overcomes the defect of the samples in existing triplet loss function algorithms being too random; learning, searching, and identifying tasks are fast and efficient, and accuracy is greatly improved.