Small sample target detection method and system based on semantic features and metric learning

Abstract

The invention discloses a small sample target detection method and system based on semantic features and metric learning, and the method comprises the steps: taking a category semantic name corresponding to a query graph as knowledge, calculating a corresponding word vector as a semantic feature by using a word2vec tool in a natural language processing field, and fusing the semantic feature with an image feature of the query graph. Semantic features are embedded into a visual domain, the distance between the targets of the same category is reduced by means of semantic consistency of the targets of the same category in a semantic space, and the distance between the targets of different categories is increased by means of semantic differences of the targets of different semantic categories in the semantic space. The problem of false detection of targets of different categories but similar appearance vision and the problem of missing detection of targets of the same semantic category but large appearance vision difference in an existing small sample target detection model based on metric learning are solved, and the detection precision of a base category and a new category is improved.

Description

technical field [0001] The invention belongs to the technical field of image detection, and in particular relates to a small sample target detection method and system based on semantic features and metric learning. Background technique [0002] The great success of deep learning models in object detection tasks is mainly due to the fact that deep neural networks can learn more advanced and deeper features from data. However, deep learning models rely heavily on a large amount of labeled data, but manual data labeling is time-consuming, laborious, expensive, and there is not enough data accumulation in some application fields. Deep learning achieves satisfactory results in data-intensive applications, but is hindered when there are few labeled data samples or small datasets. [0003] The input of a few-shot object detection model based on metric learning is a query-object image pair, and the output is the region in the object image that is similar to the query image. When t...

AI Technical Summary

Problems solved by technology

By introducing this semantic information, the distance between different types of targets can be widened and the distance between similar targets can be shortened; the problem of false detection and missed detection caused by the possibility that the appearance of similar targets may be too different and the appearance of different types of targets may be too small , improve the detection accuracy on the base class and new class data

Images