SV-UNet:A Lightweight Fracture Segmentation Model for Logging Images

The picking up of fractures in logging images is the focus of logging interpretation work. Most methods are based on mathematical statistics and traditional image segmentation, and the segmented area is not accurate enough to meet the practical needs of logging interpretation. with the development of semantic segmentation techniques proposed in other fields, it brings greater inspiration for specific target picking in logging images. In this study, when using the U-Net semantic segmentation model for fracture identification in electrical imaging logging images, poor feature utilisation and limited feature extraction capabilities of the model make it difficult to segment a single fracture in a sinusoidal shape and there is a problem of incoherent regional pixels. In order to segment the fracture region completely, this paper proposes a lightweight fracture segmentation algorithm SV-UNet, which makes full use of the advantages of U-Net and ShuffleNet V2 to enhance the network's feature transfer and utilisation and improve the feature learning capability. Experimental results on the fracture dataset of electrical imaging logging images show that the method achieves PA and mIoU of 97.18% and 78.68% respectively, which are 0.49% and 3.13% higher than those of U-Net, and has good segmentation performance.