Research On Salt Layer Segmentation Meth Od Of Geological Image Based On Deep Learning

Many areas of oil and natural gas on the earth often form a certain scale of salt deposits under the surface.In the process of oil and gas exploitation,if the position of the salt body is judged inaccurate,it will bring security risks.These salt layer information is typically reflected into the image by seismic imaging techniques.However,it is very difficult to mark the specific location of the salt body from the geological image.Because the salt layer image requires a professional to explain the salt body,this will also result in a larger subjectivity of the labeling results.In recent years,with the development of deep learning,image segmentation methods based on deep learning have also been proposed.Image segmentation based on deep learning can achieve image segmentation semantic segmentation and end-to-end processing,which is successful in many segmentation tasks.Compared with the traditional segmentation method,the deep learning method generally uses the convolutional neural network to adaptively extract features based on images and their labels,avoiding the tedious manual extraction of feature processes,modeling speed and efficient deployment.In practical applications,the performance and accuracy are significantly improved,which brings convenience to research in related fields.Since the existing deep learning image segmentation methods are all performed on specific data,such as medical images,natural images,etc.,it is difficult to directly achieve a good effect on the salt layer image,so this paper proposes a set depth based on the salt layer image.Learning image processing,segmentation,and training methods can be used to assist professionals in labeling work.In addition,image segmentation can be said to be the cornerstone of computer vision technology,and has a pivotal position in many fields.Therefore,research on image segmentation methods and performance enhancement are also important for the development of these fields.Based on the existing methods,this paper designs a geo-salt layer image segmentation method based on deep learning.In data processing,the original data is expanded by means of data enhancement.In order to further improve the generalization ability of the model,a small sample-based preprocessing method is proposed,and the model training is combined with the 5-fold cross-validation.In the network structure,this paper adopts the classic encoder-decoder structure,and uses deep network with strong classification ability in the coding stage,such as SENet,ResNet,etc.,to learn more information from the geological salt layer image.It also accelerates model convergence and improves accuracy by means of migration learning.In the decoding stage,an improved FPN network,relay supervised optimization,Hypercolumn module,full consideration of context information,fusion of multi-scale information,and further use of high-resolution and high-level features of low-level features to improve prediction accuracy.In the experimental evaluation,the threshold-based average joint cross-measure mIoU is used as the metric.Experiments show that the proposed model based on SENet,ResNet,and fusion scSE block have achieved better segmentation results,which proves that the proposed method has certain Practical value.