关键词: 深度距离度量, 深度学习, 三元组损失, 卷积神经网络, boosting

Abstract: Focused on the issue that the single deep distance metric based on triplet loss is inappropriate for handling heterogeneous data and easily lead to overfiting, a deep distance metric learning method based on optimizing triplet loss was proposed. Firstly, by thresholding the relative distance of triplet training samples mapped by neural network, and a piecewise linear discriminant function wasused as evaluation function of relative distance; Secondly, the evaluation function was added to the Boosting algorithm as weak classifier to generate a strong classifier; Finally, In order to learn the parameters of the weak classifier and neural network, an alternating optimization method wasused to obtain the optimal values. The experimental results show that the performance of proposed mehtodon the CUB-200-2011、 CARS-196 and SOP data sets is better than other methods. At the same time, the proposed method also avoids overfitting to a certain extent.

Key words: Keywords: deep distance metric, deep learning, triplet loss, convolutional neural network, boosting