Artificial intelligence identification of detrital minerals under stereoscope based on deep learning algorithm

LI Hui; HAN Zongzhu

doi:10.13984/j.cnki.cn37-1141.2025.02.006

Abstract：The identification of detrital minerals is an important part in the study of sediment mineralo-gy.Conventional identification of detrital minerals under stereoscope is a long,complex and error prone task,which must be performed by well-trained laboratory personnel.This is a tedious work,because the detrital minerals are small and lack of obvious characteristics.The fatigue of testers may lead to the reduction of work efficiency and even the wrong identification of minerals.The identification results of mineral types may be subjectively affected by testers.Computer vision and deep learning technology make it possible to identify detrital minerals intelligently.In order to improve the efficiency of detrital mineral identification and statistics,the image data set of detrital minerals under stereoscope was made,and the Yolo v3 network was trained and optimized to realize the object detection of 11 types of detrital minerals.The experiment showed that in the process of training,with the increase of training times,the loss function of the model was decreased,and finally tends was stable.The map value of the model trained with top light images and fusion images were 91.03％and 89.91％respectively,which showed that the model trained in the experiment had good object detection and recognition accuracy.Based on this,an auxiliary software for identification and statistics of detrital mineral was compiled.In the identification process of detrital mineral samples,the use of the software can reduce the workload of manual identification by 85％～90％.