Fracture identification method combining channel and spatial cross attention
The existing methods for detecting fractures in oil and gas reservoirs suffer from issues such as the use of single information,low fracture detection accuracy,slow detection speed,and a lack of training samples when applying artificial intelligence techniques.In this study,we propose a novel FCN semantic segmentation model based on attention mechanism for fracture extraction and construct a dataset for fracture recognition in well logging images.Firstly,Spatial cross attention is introduced and fused with channel attention modules in the downsampling process of the FCN model to enhance semantic information retrieval.Secondly,ResNet-50 is adopted as the backbone network of the improved model,and dilated convolutions are employed to enlarge the receptive field,thereby enhancing contextual information and improving fracture recognition accuracy.Finally,a dataset consisting of 2016 FMI well logging images is constructed,and a superpixel segmentation method is employed to assist in the manual annotation of fracture features.Comparative experiments with other classic semantic segmentation models demonstrate that our improved model achieves lower loss,higher accuracy,and an Mean-Intersection-Over-Union(MIoU)of 74.61％on the constructed FMI well logging image dataset.Additionally,the effectiveness of the proposed model is further validated through ablation experiments.The results indicate that our improved model accurately extracts fracture information from well logging images and exhibits good practicality.
NETL
NSTL
万方数据
