应用WTConvNeXt-Inception网络的岩石薄片岩性智能识别方法

打开文本图片集

关键词：WTConvNeXt-Inception网络，岩石薄片，特征融合，图像分类，深度学习中图分类号：P631 文献标识码：A DOI：10.13810/j.cnki.issn：1000-7210.20250197

Abstract： In the domains of geology and mineralogy，the precise identification ofrock thin sections holds paramountsignificance forunderstanding the composition，structure，and formation process ofrocks.However，traditionalmanual identification methods are cumbersome，time-consuming，highly subjective，and heavily dependent on experience. Deep learning enables rapid and accurate identification ofrock thin sections.This study proposes a WTConvNeXt-Inception network for inteligent lithology identification，which significantly improves clasification accuracy through multimodal feature fusion of plane-polarized and cross-polarized images. Aiming at the insuficiency of feature extraction in traditional methods when handling complex rock images，wavelet convolution is introduced to efectivelycapture the multi-scale features in the images.To address the issue of high memory consumption of the model，the Inception module is adopted to improve the operational speed and efficiency of the model.To tackle the problem of missing information in single images，across-attention fusion module is proposed， efectively exploiting the information between plane-polarized light and cross-polarized light images. Experimental results demonstrate that the proposed method achieves a classification accuracy， F1- score，and quadratic weighted Kappa values of 98.64% ， 98.64% ，and 98.30% ， respectively，verifying its effectiveness：in highly accurate and efficient identification of rock thin sections.This method shows strong potential for applications in geological research， petroleum exploration，and other related fields.

Keywords： WTConvNeXt-Inception network，rock thin section， feature fusion， image clasification，deep learning

1 概况

岩性识别在储层评价中具有重要意义，是开展油藏描述、实时钻井监测以及获取储层参数的基础工作之一。（剩余21735字）