Stratigraphic Division Method Based on the Improved YOLOv8

With the deepening of oilfield development, logging data proliferate, and their complexity makes manual stratigraphic division both difficult and time-consuming. Aimed at the current network model widely used to solve the problem of stratigraphic delineation, which has problems such as not considering the multi-scale features of logging curves and insufficient accuracy, the YOLOv8x target detection algorithm in deep learning is utilized to detect the target strata, which has the ability to characterize the multi-scale features and can improve the efficiency and accuracy of the division. In order to better localize and identify targets, this paper proposes a new stratigraphic automatic division method, YOLOv8x-CAMDP, which introduces a CA (Coordinate Attention) mechanism module into the original YOLOv8x model to improve the model's ability to identify stratigraphic interval boundaries. In addition, the CIOU loss function in the original YOLOv8x network model was replaced using the MDPIOU loss function to effectively improve the accuracy and efficiency of bounding box regression. Based on the logging data from the Xing 10 area pure oil zone, a thorough comparison of the YOLOv8x-CAMDP and YOLOv8x models' training results is presented. The YOLOv8x-CAMDP model achieves a mean Average Precision (mAP) value of 98.7%, outperforming the YOLOv8x model by one percentage point. Moreover, the YOLOv8x-CAMDP model demonstrates greater precision in boundary division for each stratigraphic interval. The application of the YOLOv8x-CAMDP model to project implementation achieved significant results in stratigraphic division, reduced workload, and optimized manual division. These results not only confirm the practical value of the YOLOv8x-CAMDP model but also demonstrate the prospect and potential of its wide application.