A Few-Sample Learning Method Based on Multi-scale Metric Learning

Abstract

The invention discloses a small-sample learning method based on multi-scale metric learning, which includes the following steps: Step 1, establishing a data set; Step 2, generating a multi-scale feature mapping layer; Step 3, transfer learning: the conversion module performs multi-scale sampling The features are remapped; step 4, generate multi-scale feature map pairs; step 5, calculate the relationship scores of multi-scale feature map pairs in the multi-scale relationship generation network; step 6, use the multi-scale metric learning model to perform sample similarity measure. The invention has a simple structure and a reasonable design. The multi-scale feature map pair is obtained through transfer learning, so that the trained model has transferability. On the basis of the mean square error loss function, the loss item brought by the sample spacing is added to the overall model to form a new model. The loss function implements metric learning to adapt to the training of small sample learning.

Description

technical field [0001] The invention belongs to the technical field of image processing and recognition, and in particular relates to a small-sample learning method based on multi-scale metric learning. Background technique [0002] Humans are very good at recognizing a new object with a very small number of samples. For example, children only need some pictures in the book to know what is a "zebra" and what is a "rhinoceros". Inspired by the rapid learning ability of human beings, researchers hope that after learning a large amount of data of a certain category, the machine learning model can quickly learn new categories with only a small number of samples. This is what Few-shot Learning is to solve question. [0003] For machine learning, although deep learning in image recognition tasks can achieve very satisfactory results in some scenarios with deep and complex network models, huge training data support, and powerful hardware support, but in In some rare task scenario...

AI Technical Summary

Problems solved by technology

However, the common algorithm of transfer learning cannot solve the small-sample learning task very well. The main difference is that the small-sample learning needs to acquire the ability to identify unknown categories, which means that in principle it has the ability to identify a large number of untrained category target, and obviously transfer learning cannot do this

Images