Image data automatic labeling method and system based on deep reinforcement learning

Abstract

The invention provides an automatic image data labeling method and system based on deep reinforcement learning and relates to the technical field of computer vision. The method comprises steps of 1, a deep learning algorithm which is used for automatically generating target detection and instance segmentation rough pre-labeling; and 2, a reinforcement learning algorithm, wherein the reinforcement learning algorithm is used for automatically correcting a labeling result and finely adjusting a rough pre-labeling result; according to the method, a rough pre-labeling result can be automatically generated by utilizing a deep learning algorithm, a correction strategy of an image labeling person on the rough pre-labeling result is learned by using a reinforcement learning algorithm, and manual intervention and correction amplitude and frequency of the labeling person are reduced through online learning of the adjustment strategy.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a method and system for automatic labeling of image data based on deep reinforcement learning. Background technique [0002] Automatic image annotation refers to the process of automatically adding textual feature information to the image to reflect its content through the method of machine learning for the visual content of the image. The basic idea is: use the labeled image set or other available information to automatically learn the potential association or mapping relationship between the semantic concept space and the visual feature space, and add text keywords to unknown images. [0003] The development of artificial intelligence image recognition algorithms usually requires a large number of labeled images for algorithm training. At present, there are mainly the following methods for image labeling: manual labeling, automatic pre-labeling based on deep le...

AI Technical Summary

Problems solved by technology

[0004] In view of the above existing technologies, the manual labeling method has the defects of long cycle time and high labor cost; while the automatic pre-labeling method based on deep learning algorithm can reduce the burden of manual labeling, but the labeling quality is lower than that of manual labeling, and in the whole In the labeling task, the quality of pre-labeling cannot be improved independently through active learning; in the method of automatic pre-labeling based on deep reinforcement learning algorithm, the algorithm can independently improve the quality of pre-labeling through active learning, but the algorithm cannot learn the manual adjustment of the labeler Therefore, it may not be able to reduce the magnitude and frequency of manual intervention and correction by annotators

Images