A method and system for recognizing paragenetic minerals based on image segmentation

By constructing a mineral image segmentation model with a multi-scale global attention encoding-decoding structure, the problem of inaccurate localization in the identification of symbiotic minerals is solved, and accurate identification and localization of symbiotic minerals are achieved, thereby improving the model's generalization ability and feature extraction ability.

CN118397611BActive Publication Date: 2026-06-05CHINA UNIV OF GEOSCIENCES (BEIJING)
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHINA UNIV OF GEOSCIENCES (BEIJING)
Filing Date
2024-05-09
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing camera image-based mineral identification methods can only identify the categories of coexisting minerals, but cannot accurately locate the position of each mineral.

Method used

A multi-scale encoder-decoder structure mineral image segmentation model with global attention is constructed. Using SwinTransformer, an improved feature pyramid, and a multi-scale pyramid pooling module, combined with a global attention module, mineral image segmentation is performed to obtain more comprehensive and richer mineral feature representations.

Benefits of technology

It achieves accurate identification and specific location of symbiotic minerals, improves the generalization ability of the mineral segmentation model, reduces the risk of overfitting, and enhances the model's ability to perceive minerals at different scales.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118397611B_ABST
    Figure CN118397611B_ABST
Patent Text Reader

Abstract

The embodiment of the application discloses a kind of based on image segmentation's paragenesis mineral identification method and system, the method and system include obtaining mineral image and marking and expanding mineral image segmentation data set, construct with Swin Transformer as encoder, improved feature pyramid as decoder mineral image segmentation model, and using multi-scale pyramid pooling and global attention module to fuse the mineral feature information of different scales, obtain more comprehensive and rich mineral feature representation, effectively integrate the context information of mineral image, strengthen the extraction ability of model for different abstract level features, maintain the richness of feature map, improve mineral segmentation performance, when training, using transfer learning to freeze the weight of the encoder Swin Transformer after pre-training in ImageNet and fine-tuning in mineral classification data set, the model obtained using mineral segmentation data set is trained, the obtained model solves the problem that existing method and system can only identify mineral species in paragenesis mineral image, cannot accurately locate various mineral.
Need to check novelty before this filing date? Find Prior Art