A Few-Shot Object Classification Method Based on Two-way Structural Contrastive Embedding Learning

Abstract

The present invention relates to the field of computer vision, in particular to a small-sample object classification method based on two-way structure contrast embedding learning, comprising the following steps: Step 1: using large-scale labeled data of basic classes to construct a two-way structure contrast embedding network; step Two: Input a small number of labeled pictures of the new class into the two-way structure comparison embedding network in turn, extract the corresponding picture features, and calculate the mean value of all picture features in each category as the prototype of the category; Step 3: Input the test picture The two-way structure contrasts the embedded network to extract image features, calculates the cosine similarity between the test image features and all categories of prototypes, and normalizes the cosine similarity, and selects the object category with the maximum similarity as the test image. The final prediction result. The method of the invention is simple and flexible, and can improve the generalization ability of the feature embedding learning network, thereby increasing the training speed of the model and improving the classification performance of new class objects.

Description

technical field

[0001] The invention relates to the field of computer vision, in particular to a small-sample object classification method based on two-way structure contrast embedding learning. Background technique

[0002] In recent years, deep neural network technology has made remarkable progress in a variety of computer vision tasks, and one of the success factors lies in the use of large-scale labeled datasets. However, in many practical scenarios, we usually only have a small amount of labeled data, such as medical image processing, industrial manufacturing, etc. When directly using deep neural networks to solve the above practical problems, it will lead to serious overfitting and model bias. , causing catastrophic performance degradation. In contrast, even preschool children can quickly learn and understand new object concepts after seeing only a few pictures. In order to narrow the ability gap between intelligent machines and human learning, the research on machin...

Images