Search

CN-121213591-B - Image annotation boundary automatic correction method and system based on semi-supervised learning

CN121213591BCN 121213591 BCN121213591 BCN 121213591BCN-121213591-B

Abstract

The invention relates to the technical field of image recognition, in particular to an automatic correction method and an automatic correction system for an image labeling boundary based on semi-supervised learning, wherein the system comprises an image processing module, a detection module and a detection module, wherein the image processing module comprises an acquisition unit for respectively acquiring an image to be corrected and a labeling boundary characteristic; the model training module comprises a model training unit used for training an initial model according to the marked boundary features, a boundary correction module used for detecting and automatically correcting the boundary features to be corrected according to a semi-supervised learning model, a matching degree adjusting module used for determining the matching degree of the time features of the images to be corrected according to the acquired clock offset of the images to be corrected, a pseudo-tag adjusting module used for determining the iterative evolution rate of the pseudo-tag according to the increase rate of the boundary errors of a test set in a prediction model, and a learning rate adjusting module used for determining the noise robustness learning rate scaling factor according to the noise occupation ratio in the pseudo-tag. The invention improves the correction stability of the image labeling boundary.

Inventors

  • GUAN RUI
  • SUN YU

Assignees

  • 北京智睿博科技有限公司

Dates

Publication Date
20260505
Application Date
20250922

Claims (9)

  1. 1. An automatic correction system for an image annotation boundary based on semi-supervised learning is characterized by comprising: the image processing module comprises an acquisition unit for respectively acquiring an image to be corrected, marked boundary features and unmarked boundary features and a preprocessing unit connected with the acquisition unit and used for preprocessing the image to be corrected to output the boundary features to be corrected; The model training module is connected with the image processing module and comprises a model training unit, a pseudo tag generating unit and a model updating unit, wherein the model training unit is used for training an initial model according to the marked boundary features to output a prediction model, the pseudo tag generating unit is connected with the model training unit and used for predicting the unmarked boundary features according to the prediction model to generate a pseudo tag, and the model updating unit is connected with the pseudo tag generating unit and used for updating the prediction model according to the pseudo tag and the marked boundary features to obtain a semi-supervised learning model; the boundary correction module is connected with the model training module and used for detecting and automatically correcting the boundary characteristics to be corrected according to the semi-supervised learning model; the matching degree adjusting module is connected with the image processing module and used for determining the matching degree of the time characteristics of the image to be corrected according to the acquisition clock offset of the image to be corrected; The pseudo tag adjusting module is respectively connected with the model training module and the matching degree adjusting module and is used for determining the iterative evolution rate of the pseudo tag according to the boundary error increasing rate of the test set in the prediction model; the learning rate adjusting module is respectively connected with the model training module and the pseudo tag adjusting module and used for determining a noise robustness learning rate scaling factor according to the noise duty ratio in the pseudo tag; the matching degree of the time characteristics of the images to be corrected is the consistency degree of the boundary marking characteristics of the images to be corrected of the same style acquired at different times; The iteration evolution rate of the pseudo tag is the mass change speed of the pseudo tag along with the iteration of the prediction model in the semi-supervised training iteration process; The noise robustness learning rate scaling factor is a coefficient for dynamically adjusting the learning rate weight according to the noise degree of the data; the matching degree adjusting module is used for determining that the correction stability of the image marking boundary meets the requirement in response to the fact that the acquired clock offset of the image to be corrected is smaller than or equal to a preset first offset; and the matching degree adjusting module determines that the correction stability of the image annotation boundary does not meet the requirement in response to the acquisition clock offset of the image to be corrected being greater than the preset first offset.
  2. 2. The system for automatically correcting the image annotation boundary based on semi-supervised learning as recited in claim 1, wherein the matching degree adjustment module initially determines that the generalization of the prediction model is not satisfactory in response to the acquired clock offset of the image to be corrected being greater than the preset first offset and less than or equal to the preset second offset.
  3. 3. The automatic correction system of image annotation boundaries based on semi-supervised learning according to claim 2, wherein the matching degree adjustment module increases the matching degree of the time feature of the image to be corrected in response to the acquired clock offset of the image to be corrected being greater than the preset second offset; the increasing amplitude of the matching degree of the time characteristics of the image to be corrected is determined by the difference value between the acquired clock offset of the image to be corrected and the preset second offset.
  4. 4. The semi-supervised learning based image annotation boundary automatic correction system of claim 3, wherein the pseudo tag adjustment module determines that generalization of the predictive model meets requirements in response to the test set boundary error growth rate in the predictive model being less than or equal to a preset first growth rate; and the pseudo tag adjusting module determines that the generalization of the prediction model is not in accordance with the requirement in response to the fact that the boundary error growth rate of the test set in the prediction model is larger than the preset first growth rate.
  5. 5. The semi-supervised learning based image annotation boundary automatic correction system of claim 4, wherein the pseudo tag adjustment module reduces the iterative evolution rate of pseudo tags in response to the test set boundary error growth rate in the predictive model being greater than the preset first growth rate and less than the preset second growth rate; And the pseudo tag adjusting module initially determines that the boundary noise immunity of the prediction model is not satisfactory in response to the test set boundary error growth rate in the prediction model being greater than the preset second growth rate.
  6. 6. The system for automatically correcting the boundary of the image annotation based on semi-supervised learning as recited in claim 5, wherein the magnitude of the decrease in the iterative evolutionary rate of the pseudo tag is determined by the difference between the test set boundary error growth rate and the preset first growth rate in the predictive model.
  7. 7. The automatic correction system of image annotation boundaries based on semi-supervised learning of claim 6, wherein the learning rate adjustment module determines that the boundary noise immunity of the predictive model meets the requirements in response to the noise duty ratio in the pseudo tag being less than or equal to a preset noise duty ratio; And the learning rate adjusting module determines that the boundary noise immunity of the prediction model is not in accordance with the requirement and increases the noise robustness learning rate scaling factor in response to the noise duty ratio in the pseudo tag being greater than the preset noise duty ratio.
  8. 8. The semi-supervised learning based image annotation boundary automatic correction system as recited in claim 7, wherein the magnitude of the increase in noise robustness learning rate scaling factor is determined by the difference between the noise duty cycle in the pseudo tag and a preset noise duty cycle.
  9. 9. A correction method applied to the semi-supervised learning-based image annotation boundary automatic correction system as recited in any one of claims 1-8, comprising: respectively collecting an image to be corrected, marked boundary features and unmarked boundary features, and sequentially carrying out cleaning, denoising, conversion and feature extraction on the image to be corrected to output the boundary features to be corrected; training an initial model by using the marked boundary features to obtain a prediction model, predicting the unmarked boundary features according to the prediction model to generate a pseudo tag, updating the prediction model according to the pseudo tag and the marked boundary features to obtain a semi-supervised learning model, and detecting and automatically correcting the boundary features to be corrected according to the semi-supervised learning model; Acquiring an acquisition clock offset of an image to be corrected, and determining whether the correction stability of an image marking boundary meets the requirement or not according to the acquisition clock offset of the image to be corrected; If the correction stability of the image annotation boundary does not meet the requirement, determining whether the matching degree of the time characteristic of the image to be corrected needs to be increased; If the matching degree of the time characteristics of the image to be corrected does not need to be increased, acquiring the boundary error growth rate of the test set in the prediction model to determine whether the generalization of the prediction model meets the requirement; if the generalization of the prediction model does not meet the requirements, determining whether the iterative evolution rate of the pseudo tag needs to be reduced or not; if the iterative evolution rate of the pseudo tag does not need to be reduced, a noise robust learning rate scaling factor is determined based on the noise duty cycle in the pseudo tag.

Description

Image annotation boundary automatic correction method and system based on semi-supervised learning Technical Field The invention relates to the technical field of image recognition, in particular to an automatic correction method and system for an image annotation boundary based on semi-supervised learning. Background In the field of modern computer vision, an image annotation boundary automatic correction technology based on semi-supervised learning has become a key technical support for guaranteeing the quality of annotation data by virtue of the core advantages of low annotation data dependence, high correction efficiency and strong deviation adaptability. The correction accuracy directly determines the reliability of the labeling data and the training effect of the models such as target detection, image segmentation and the like. Along with the development of the automatic image annotation boundary correction technology based on semi-supervised learning to scene generalization and precision self-adaptation, the data related to the annotation correction process and scene characteristics show the characteristics of multisource, high frequency and high heterogeneity, and the difference of the requirements of different types of data on real-time correction response, deviation recognition precision and cross-scene compatibility is obvious. Under the background, the traditional automatic correction mode of the image annotation boundary based on semi-supervised learning gradually exposes a plurality of technical bottlenecks, and is difficult to meet the requirement of accurate annotation in complex scenes. CN118608554A discloses an ultrasonic medical image segmentation method based on semi-supervised learning, which comprises the following steps of training an image restoration model based on a first image dataset and a degradation image dataset obtained by downsampling the first image dataset, initializing a first pre-training model and a second pre-training model based on the image restoration model, training the first pre-training model based on the second image dataset carrying a real segmentation label to obtain a first segmentation model, inputting a third image dataset into the first segmentation model to obtain a pseudo segmentation label of the third image dataset output by the first segmentation model, and training the second pre-training model based on the third image dataset carrying the pseudo segmentation label and a fourth image dataset carrying the real segmentation label to obtain an image segmentation model. Therefore, the ultrasonic medical image segmentation method based on semi-supervised learning has the problem that the correction stability of the image labeling boundary is insufficient due to the fact that the confidence threshold is not controlled when the pseudo-segmentation labels are screened, and the model learning error labeling information is caused by the fact that the number of the noise pseudo-labels is too large. Disclosure of Invention Therefore, the invention provides an automatic correction method and an automatic correction system for an image labeling boundary based on semi-supervised learning, which are used for solving the problem that the correction stability of the image labeling boundary is insufficient due to the fact that the number of noise pseudo labels is too large and the model learning error labeling information is caused by the fact that a confidence threshold is not controlled when the pseudo segmentation labels are screened in the prior art. In order to achieve the above object, the present invention provides a method and a system for automatically correcting an image labeling boundary based on semi-supervised learning, comprising: the image processing module comprises an acquisition unit for respectively acquiring an image to be corrected, marked boundary features and unmarked boundary features and a preprocessing unit connected with the acquisition unit and used for preprocessing the image to be corrected to output the boundary features to be corrected; The model training module is connected with the image processing module and comprises a model training unit, a pseudo tag generating unit and a model updating unit, wherein the model training unit is used for training an initial model according to the marked boundary features to output a prediction model, the pseudo tag generating unit is connected with the model training unit and used for predicting the unmarked boundary features according to the prediction model to generate a pseudo tag, and the model updating unit is connected with the pseudo tag generating unit and used for updating the prediction model according to the pseudo tag and the marked boundary features to obtain a semi-supervised learning model; the boundary correction module is connected with the model training module and used for detecting and automatically correcting the boundary characteristics to be corrected according to the semi-supe