基于YOLOv8神经网络的岩芯结构面智能识别技术OA

[Objective]Image recognition is an important technology in the field of artificial intelligence.It realizes image recognition by processing images via computers,extracting image features,and identifying key information.At present,discontinuities in rock cores in engineering investigation are mainly identified manually,which greatly restricts logging efficiency.Therefore,a target detection algorithm based on the YOLOv8 neural network is proposed for borehole rock core images to achieve intelligent recognition of characteristic discontinuities such as soil-rock boundaries,joint planes,and weathering interfaces.[Methods]First,rock core images from railway projects such as the Changsha-Ganzhou and Wenzhou-Fuzhou railways were collected to construct a dataset of rock core discontinuities.The dataset contained more than 30 000 labeled instances,and the Mosaic data augmentation algorithm was used to expand the labeled data.Then,an intelligent recognition model for rock core discontinuity information was constructed based on the YOLOv8 neural network.The model was trained,and its recognition accuracy was improved through transfer learning and continuous model tuning.Next,model evaluation indicators were proposed,and the model was evaluated based on its recall and precision.[Results]The intelligent recognition model for borehole rock core images achieved an overall recognition accuracy of 86.6%,with a recognition speed of 7.28 s per image.Compared with traditional manual identification,the recognition efficiency was improved by more than four times.[Conclusion]The constructed intelligent recognition model exhibits good applicability to engineering sites,achieves relatively high recognition accuracy for rock core discontinuities,and can significantly improve recognition efficiency.