A Design Method of Triplet Loss Function Based on Trace Ratio Criterion

Abstract

A triplet loss function design method based on the trace ratio criterion, through the investigation of image feature extraction, triplet loss function and trace ratio criterion, using the trace ratio criterion (Trace Ratio Criterion) as the triplet selection criterion and loss Calculation method. The method mainly includes: A, triplet sample construction: for each sample in the data set, construct it into a triplet sample; B, triplet sample selection: screen from the constructed triplet sample, Set an effective selection mechanism to improve the training speed without losing accuracy; C. Loss function design: According to the triplet samples obtained in B, calculate the distance between the current sample and the positive and negative samples in the triplet respectively , design a loss function to calculate the error between the model prediction and the real result; D, deep network training: return the model error to the deep convolutional neural network, update and adjust the network parameters, and iteratively train the model until convergence.

Description

technical field [0001] The invention belongs to the field of loss function design, and in particular relates to a triplet loss function design method based on the trace ratio criterion. The method uses processed image data as the most original input. Background technique [0002] In recent years, with the rise of deep learning, computer vision technology based on convolutional neural networks has developed rapidly. The convolutional neural network is a common deep neural network, and its origin can be traced back to the proposal of the backpropagation algorithm. A conventional neural network consists of an input layer, a hidden layer, and an output layer. Each hidden layer contains several neurons, and each neuron is fully connected to the neurons in the previous layer. The convolutional neural network is different from the conventional neural network. For the case where the input is an image, the neurons in each layer of the network are arranged in three dimensions (width,...

AI Technical Summary

Problems solved by technology

This method uses a fixed parameter α throughout the training process, which cannot achieve the adaptability of the parameters at each iteration, making the generalization ability of the final model insufficient.

Images