Abstract: Aiming at addressing the issue of serious inefficiency in the traditional manual evaluation methods for rock core integrity,an intelligent rock core integrity evaluation algorithm,named YOLOv8-CMR (YOLOv8 enhanced by Contextual dynamic information and Multi-scale attention for Rock core integrity prediction),is proposed.In YOLOv8-CMR,a dynamic feature extraction module based on context enhancement is firstly constructed by fusing ODConv (Omni-Dimensional Dynamic Convolution) and Transformer.ODConv is used to realize accurate extraction of complex and variable core detail features by all-round dynamic adjustment of convolutional kernel parameters,and Transformer is used to extract the global context information by analyzing the deep feature layer to enhance the network’s anti-interference ability.Then,a multi-scale attention-based feature enhancement module is designed to improve the multi-scale feature extraction capability by assigning weights to the feature layers of different scales in Atrous Spatial Pyramid Pooling.After that,the fused feature layer is decoded by the feature decoding module to output the rock core’s detection results.Finally,Rock Quality Designation (RQD) is automatically calculated and the rock core integrity level is predicted according to the detection results.The experimental results show that F1 and mPA of YOLOv8-CMR are 93.9% and 95.6%,respectively,which are superior to the four commonly used recognition algorithms including DETR,Faster-RCNN,EfficientDet,and YOLOv8.In addition,the RQD prediction result of YOLOv8-CMR is 2.08% error from manual measurement,and the prediction accuracy of rock core integrity grade is 93.3%,which is suitable for intelligent evaluation of rock core integrity.