Abstract: Rock Quality Designation (RQD),as a key indicator for evaluating the integrity of rock masses,is widely used
in geological and mining engineering,providing important basis for engineering design and construction. Traditional RQD determination
methods rely on manual measurement,which suffer from issues such as low acquisition efficiency and large error in results.
To address this,an intelligent RQD calculation method based on image detection and object segmentation is proposed.
This method first constructs a rock core image dataset through perspective transformation,data augmentation,and image annotation
techniques. Subsequently,structural improvements are made to the Segment Any Model (SAM) in two aspects:firstly,an
Adapter module is introduced into the image decoder for fine-tuning to enhance the model′s representation ability for rock core
features;secondly,the loss function is optimized to improve the segmentation accuracy of rock core edges and achieve precise
extraction of individual rock core segments. Finally,to address the issue of rock core inclination,the Hough transform is used
for pose correction,and the rock core length is measured using the median line-pixel statistical method to complete the RQD
calculation. Experimental results show that the improved SAM model yields clear rock core boundaries and complete contours,
with an F1 value of 95. 21% and an Intersection over Union (IoU) of 88. 91%. The average absolute error between the intelligent
RQD calculation results of rock cores and traditional manual calculation results is no more than 5%,indicating high accuracy.
Compared to manual measurement,the intelligent RQD calculation of rock cores significantly reduce the time consumption and effectively improve the efficiency of RQD calculation.

Key words: core image segmentation,SAM model,perspective transformation,adapter fine-tuning,Hough transform,rock
quality designation