Intelligent Recognition Model of Rock Samples

Abstract

At present, the methods of rock sample identification mainly include remagnetization, well logging, seismic, remote sensing, electromagnetism,geochemistry, hand specimen and thin section analysis, etc. However, these lithologic classification methods are susceptible to various conditions such as expert’s capacity, weather, climate and geographical differences, etc., and they also require certain human and material resources. In order to realize the classification recognition of rock lithology, this paper innovates the contour extraction detection algorithm,named DAYOLO ,which is mainly based on YOLOv3 network structure.We added the steps of contour extraction and used mean filtering to remove noise. After the grayscale conversion of the image, we use the Lagrangian sharpening operator to extract the contour of the object. In order to further improve the model, we use the density clustering algorithm (DBSCAN) to test abnormal contour points, and use the artificial neural network (ANN) to fit the contour and fill in the missing values, so as to achieve the effect of contour extraction.On this basis, we use the YOLO framework to get a higher classification accuracy. The experimental results show that our algorithm is suitable for a variety of target detection models, and it can accurately propose the outline of the rock sample and complete the task of lithology recognition.