基于MEGNet的岩心孔洞图像分割算法OA

In view of the high similarity between foreground and background,large difference in hole morphology and blurred edge segmentation in core hole segmentation,this paper proposes an edge-guided segmentation model MEGNet based on Mamba.Firstly,the parallel PVM Block is designed as the backbone module to model the long-range dependence,which reduces the number of parameters and alleviates the missegmentation.Secondly,an edge generation(EG)module is constructed to generate edge features by fusing low-level details and high-level semantic features.Then,an edge-guided attention(EGA)module is proposed to optimize edge details by combining reverse features and multi-scale channel attention module(MS-CAM).Finally,the feature enhancement module(FEM)is introduced,and the multi-scale context information is captured by multi-expansion rate dilated convolution to enhance the expression of key features.Experiments show that the F1-score,intersection over union(IoU)and mean intersection over union(MIoU)of MEGNet on the core hole dataset reach 88.83%,79.92%and 89.73%,respectively.The proposed method has better segmentation effect and excellent performance in comparison with the mainstream semantic segmentation models.