A Salt Body Recognition Method Based on Deep Learning Semantic Boundary Enhancement

Abstract

The invention discloses a salt body recognition method based on deep learning semantic boundary enhancement. In addition to focusing on the recognition function of semantic images, the invention also trains the extraction of semantic boundaries to enhance the recognition effect of semantic images. Since the network has obtained the relevant ability of boundary extraction, the boundary of the semantic image output by the model will become clearer and the accuracy rate will be increased. Moreover, the features of semantic boundary recognition will also be directly and explicitly input into the process of semantic image extraction, so as to directly supervise and strengthen the results of salt body recognition. The attention module scSE in the semantic image extraction network also allows the model to learn by itself during the training process, obtain the importance of each feature, and then improve useful features and suppress features that are not very useful for the current task according to this importance. Modeling the interdependence between feature channels formally also makes the model more stable. The method can efficiently and accurately segment the geological salt layer image.

Description

technical field [0001] The invention belongs to the technical field of computer vision and the field of geological survey. The main knowledge involved includes some image enhancement, boundary detection, image semantic segmentation methods, deep learning image segmentation methods, etc. Background technique [0002] Seismic image refers to the observation of the velocity of seismic waves propagating under different rock layers through seismic imaging technology, and then obtains the corresponding acoustic wave map. Seismic imagery, as an organized and understandable display of data, is an invaluable tool for obtaining and communicating information about the Earth's structure and material properties. Usually near huge salt deposits underground, there are often large amounts of important resources such as oil or natural gas. Areas where these resources accumulate tend to form bodies of salt below the surface. However, salt bodies usually exist in the form of high-temperatur...

AI Technical Summary

Problems solved by technology

First of all, the seismic image itself with good data annotation is not rich in data, which is not conducive to the training of the network

At the same time, due to its imaging principle, the difference between the salt body area and other background areas is small, and there is a lack of clear semantic boundaries, which will make the segmentation results of the model inaccurate at the boundaries

Images