Semi-supervised image semantic segmentation method based on entropy minimization

Abstract

The invention relates to a semi-supervised image semantic segmentation method based on entropy minimization. The method comprises the steps: providing a feature gradient mapping regularization strategy (FGMR), wherein the gradient mapping of a low-layer feature map in an encoder is used for enhancing the encoding capacity of the encoder for a deep-layer feature map; providing a self-adaptive sharpening strategy, and keeping the decision boundary of the unmarked data in a low-density region; and in order to further reduce the influence of noise, proposing a low-confidence consistency strategy to ensure the consistency of classification and segmentation. A large number of experiments prove the superiority of the algorithm compared with the existing method.

Description

technical field [0001] The invention belongs to the technical field of computer vision and is used for semi-supervised image semantic segmentation. It plays a vital role in image segmentation in the scene of only a small amount of labeled data and a large amount of unlabeled data. Supervised Image Semantic Segmentation Methods. Background technique [0002] In recent years, with the development of deep supervised learning, various computer vision tasks have achieved remarkable progress. However, training a deep neural network requires a large amount of labeled data, which is usually time-consuming and expensive to obtain. Especially in semantic segmentation tasks, a large number of pixel-level labels are required, and the annotation cost is 15 times and 60 times that of region-level and image-level labels, respectively. Due to the need for annotation by professional doctors, the increase in the cost of medical image segmentation is more obvious. Therefore, more and more a...

AI Technical Summary

Problems solved by technology

For example, GAN-based methods utilize unlabeled data, but require careful design of specific network structures and are difficult to train

The method of consistency training requires forward propagation of each perturbed data separately, resulting in additional calculations, and the perturbation implicitly enhances the data. For a fully supervised network model without data enhancement, the difference between the two The performance comparison is unfair

Images