Coronavirus X-ray image recognition method based on improved MobileNetV3

By using an improved MobileNetV3 network and a weighted channel screening module, the accuracy and efficiency of X-ray image recognition for COVID-19 have been enhanced, solving the misdiagnosis problem of existing recognition models and enabling rapid and accurate case screening.

CN115410027BActive Publication Date: 2026-06-09GUANGDONG UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
GUANGDONG UNIV OF TECH
Filing Date
2022-07-18
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing technologies for COVID-19 X-ray image recognition models struggle to balance high accuracy and efficiency, especially in large-scale case screening where there is a risk of misdiagnosis.

Method used

An improved MobileNetV3 network is adopted, combined with a weighted channel filtering module. Through attention mechanism and weighted channel discarding method, the channel features of the feature map are enhanced, discriminative features are filtered out and redundant information is removed. Combined with the lightweight network MobileNetV3-small, fast recognition is achieved.

Benefits of technology

It improved the accuracy and efficiency of X-ray image recognition for COVID-19, reduced the risk of misdiagnosis, and met the real-time requirements for large-scale case screening.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115410027B_ABST
    Figure CN115410027B_ABST
Patent Text Reader

Abstract

The application discloses a COVID-19 X-ray image recognition method based on an improved MobileNetV3, and comprises the following steps: obtaining an X-ray image to be recognized, inputting the X-ray image into a trained COVID-19 image recognition model, and obtaining a category to which the X-ray image belongs; the COVID-19 image recognition model adopts a lightweight network MobileNetV3-small as a backbone network, and a weighted channel screening module is constructed, based on an attention mechanism and a weighted channel dropout method, so that feature enhancement of an input feature map in a channel is realized; the weighted channel screening module is combined with the backbone network, the weighted channel screening module firstly extracts high-dimensional and low-dimensional channel feature weights of the input feature map from the MobileNetV3-small, and then reallocates scores of each feature channel; then, a weighted random sampling strategy is adopted to generate a mask of the high-dimensional and low-dimensional feature channels, the high-dimensional and low-dimensional weights are fused, and the mask is used for channel screening of the fused weights; finally, the fused weights are assigned to the input feature map.
Need to check novelty before this filing date? Find Prior Art