Multi-label image classification method based on direct push type semi-supervised deep learning

Abstract

The invention discloses a multi-label image classification method based on direct push type semi-supervised deep learning, and the method specifically comprises the steps: 1) preparing a multi-label image training sample set, and dividing the training sample set into a labeled training sample set and an unlabeled training sample set; 2) constructing a class symbiosis relation graph; 3) building a network model according to the class symbiosis relation graph; 4) constructing a cross-correlation entropy classification loss function with confidence; 5) constructing a classification probability constraint function with confidence; 6) constructing a category-feature correlation objective function with confidence; 7) constructing a total objective function; (8) training the network model built in the step (3) according to the total objective function constructed in the step (7) in combination with a direct push type semi-supervised learning principle; and 9) inputting a to-be-classified multi-label image into the network model trained in the step 8), and obtaining a prediction category of the image in a classification layer. According to the method, the feature vectors with higher category correlation can be learned.

Description

technical field [0001] The invention belongs to the technical field of computer vision multi-label image classification, in particular to a multi-label image classification method based on transductive semi-supervised deep learning. Background technique [0002] Multi-label image refers to an object with multiple class labels in an image, and multi-label image classification refers to assigning multiple class labels related to each image from a predefined set of class labels. Multi-label image classification has a wide range of application requirements. The application scenarios of multi-label image classification include multi-target recognition and detection, album classification, image semantic annotation, image retrieval, etc. [0003] Supervised deep learning methods driven by big data have achieved state-of-the-art performance in multiple computer vision application fields such as multi-label image classification. However, constructing a large-scale annotated trainin...

AI Technical Summary

Problems solved by technology

[0007] In summary, for multi-label image classification tasks, the current semi-supervised learning methods still have certain limitations and deficiencies.

Images