A CT scan image mineral intelligent segmentation method based on SESUNet

By using a segmentation method based on SESUNet, the accuracy and robustness issues of traditional algorithms in mineral particle segmentation are solved, and automatic and accurate segmentation of mineral particle images is achieved, which is suitable for real-time analysis and resource assessment in mines.

CN121505615BActive Publication Date: 2026-07-03CHINA UNIV OF MINING & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHINA UNIV OF MINING & TECH
Filing Date
2025-11-18
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Traditional image segmentation algorithms struggle to achieve ideal results in mineral particle segmentation tasks, especially when mineral particles are uneven in size, adherent, have blurred edges, and are irregular in shape. They are unable to achieve accurate segmentation results and lack self-learning capabilities, relying on operator experience to adjust parameters, thus failing to adapt to the imaging characteristics of different types of mineral samples.

Method used

A segmentation method based on SESUNet is adopted. By constructing a mineral CT image dataset, feature extraction and restoration are performed using SCConv and SENet modules. The model is trained by combining a pre-computed weighted cross-entropy loss function to optimize the model and improve segmentation accuracy.

Benefits of technology

It enables automatic and accurate segmentation of mineral particle images, improves the accuracy of edge detection and the precision of boundary segmentation of adherent mineral particles, meets the real-time analysis needs of mine sites, and provides accurate and reliable basis for mineral particle analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121505615B_ABST
    Figure CN121505615B_ABST
Patent Text Reader

Abstract

A CT scanning image mineral intelligent segmentation method based on SESUNet, comprising the following steps: constructing a mineral CT image dataset; constructing a segmentation model based on SESUNet; and training and optimizing the constructed segmentation model based on SESUNet.The weighted cross-entropy loss function is used to replace the standard loss function in the network optimization stage, higher weight is assigned to the foreground mineral pixels, the class difference is effectively balanced, the model pays more attention to the mineral area difficult to identify in the training, and the segmentation precision of the model and the recall rate of rare minerals are directly improved from the optimization target level.The present application realizes the automatic and accurate segmentation of mineral particle images, and provides accurate and reliable basis for subsequent mineral particle analysis and resource evaluation.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • CN118429651A

  • CN119919429A