| Mineral recognition and discrimination play a significant role in geological study.In this study,the principles and characteristics of some common mineral analysis technologies were summarized to promote a more efficient mineral images intelligent discrimination technology: Characteristics of quartz,biotite and K-feldspar from granite thin sections under cross-polarized light were studied for mineral images intelligent classification by Inception-v3 CNN and transfer learning method.30 mineral particles from each type of mineral: quartz,biotite and K-feldspar in granite thin sections were selected.And their dynamic images at 16 angles from 0° to 90° with equal interval under cross-polarized light were used to establish our mineral images dataset.Then,these images were randomly divided into training set and validation set according to the ratio of 8:2 for the training and validation of Inception-v3 CNN in the process of transfer learning.The final accuracy of the well-trained model on the validation set reaches 97.00%.And model test results show that the average discrimination accuracies of quartz,biotite and K-feldspar are 100.00%,96.88% and 90.63%.During the application phase,the well-trained Inception-v3 CNN can correctly discriminate quartz,biotite and K-feldspar in different granite thin sections,and their average discrimination accuracy is 93.75%,81.25%and 75% respectively.Results of this study prove the feasibility and reliability of the application of convolution neural network in mineral images classification,and shows that using multi angle images of mineral particle under cross-polarized light can significantly improve the accuracy of mineral discrimination.This study represents a step forward compared to previous studies by bringing the mineral images intelligent classification from static images classification to dynamic multi-angles images classification.And it could provide a new perspective for the development of more professional and practical mineral intelligent discrimination applications. |
