基于改进YOLOv5s实现岩石薄片中副矿物实时视觉识别与定位OA北大核心

In early research on mineral identification which used microscopes and were based on deep learning,the main focus was on the identification of major minerals with large grain sizes.As such,there is currently a dearth of studies pertaining to the identification of significant but smaller accessory minerals.In this paper,we propose a method for identifying zircon,an accessory mineral,in three principal rock classes using the improved YOLOv5s algorithm.The YOLOv5 framework is enhanced by the integration of the improved efficient channel attention and simplified spatial pyramid pooling-fast modules.After training the model using zircon data from thin sections,the obtained optimal training weight file is combined with libraries such as MSS(multiple screen shots)to achieve the real-time identification and distribution of zircon in rock samples.The results show that the proposed method improves the precision,recall rate,and mAP50 of the improved YOLOv5s model to 88%,82%,and 86%,respectively,compared with previous methods.Therefore,combined with a rich dataset,the improved YOLOv5s model proposed in this paper can achieve real-time and accurate identification of small minerals in thin sections.