Improved DeeplabV3+algorithm for fracture extraction in core images
Core fractures are of great significance for oil and gas exploration and serve as valuable geological research data.Extracting fractures from core fracture images helps geological experts in subsequent research work.However,core fracture images face challenges such as tiny fractures and pixels in the fracture and background areas being close in value,leading to unsatisfactory results from existing image segmentation algorithms.To improve the extraction of core fractures,this paper proposes an algorithm based on an improved DeeplabV3+for fracture extraction from core images.The proposed algorithm designs a new decoder that fully performs multi-scale feature fusion on the image,enhancing the model's ability to capture detailed fracture edges.Additionally,a Strip Pooling Module(SPM)is introduced as the pooling layer in the Atrous Spatial Pyramid Pooling(ASPP),effectively reducing interference from the background areas during fracture target extraction.Experimental results demonstrate that the proposed Improved DeeplabV3+Algorithm for Fracture Extraction in Core Images shows good performance in extracting fractures from core fracture images,with enhancements of 1.88%,4.49%,and 3.02%in mIoU,mPA,and F1 Score,respectively,compared to the original DeeplabV3+network.
万方数据
维普
