Boosting vision transformer for low-resolution borehole image stitching through algebraic multigrid

In geotechnical engineering, borehole image stitching technology is needed to detect cracks in borehole walls and quicksand, etc., which helps prevent natural disasters. However, traditional manual stitching of low-resolution borehole images requires a lot of manual work by experts, which leads to very expensive costs. In this paper, We propose Vision Transformer for low-resolution borehole image mosaic framework integrating algebraic multigrid (AMG), which saves a lot of manual costs and gradually moves towards automation. Then, we design AMG enhanced Vision Transformer stitching model for borehole image, which overcomes the problems of stitching seam and black image blocks caused by image distortion and blur. Finally, we collect a large number of real data sets from different engineering sites, such as natural gas tunnel, oil exploration and mining ore detection, so that our proposed method can better adapt to different geotechnical environments. Experiments show that our method has better monotonicity, and the result of the stitching image is more complete than the existing results based on SIFT, APAP and COTR.