Multi-task Triplet loss function learning method based on semantic hierarchy

Abstract

The invention provides a multi-task Triplet loss function learning method based on a semantic hierarchy. The multi-task Triplet loss function learning method comprises the following steps: constructing a semantic hierarchy network for a database, performing triplets sampling on the semantic hierarchy, training the multi-task Triplet network, and performing multi-task classification by using a treeclassifier. The multi-task Triplet loss function learning method proposes a loss function combining a semantic hierarchy network with Triplet for solving the problem of multi-level training of the Triplet network, utilizes semantic knowledge to guide a network to differentiate sample structures hierarchically, learns a Triplet feature which contains semantic hierarchy information and is higher ingeneralization, effectively applies the Triplet feature to multi-task learning, and improves feature separability under different semantic hierarchies. Meanwhile, a new hierarchical sampling method is studied, so that the network can mine more effective hard triplets, and finally, the performance of the network is improved.

Description

technical field [0001] The invention relates to the field of machine learning and image recognition, in particular to the learning of multi-task Triplet loss function based on semantic hierarchy. Background technique [0002] With the increase of convolutional neural network calculation and task complexity, traditional learning methods have shown many shortcomings, and metric learning, which updates network parameters by calculating the similarity between images, has become an important method in this field. A method for efficiently retrieving similar images. The purpose of metric learning is to calculate the similarity between images, separate heterogeneous images and aggregate similar images in the feature space, and finally learn image features with strong distinguishability. The goal of the traditional softmax loss function is to compress the data features of different categories into a certain range of the feature space, so the differences between classes are not prese...

AI Technical Summary

Problems solved by technology

But the Triplet network also has shortcomings. For example, the loss function of the network converges slowly, and the sampling method has too much influence on the network. These defects make the network largely dependent on the availability of the selected triplets.

Therefore, training Triplet networks in fine-grained images is a very challenging task

Images