AI researchers at Deepkapha propose “VeerNet,” a new switch-based deep learning network that uses self-attention processes to learn distinct curves from a single path.

Well logging is the process of using drilling tools to take measurements of different rock characteristics along a well down to the ground. The lithology of composition, porosity, fluid content, and compositional variability determine the responses of the digital record. Well notation parameters generate stony groups and descriptors of soft rock properties one by one. Prior to the introduction of digital logging tools, well logging data was plotted in a curve format on the parameter graph. Good recording parameter graphs have many drawbacks, including their large size, abundant memory space, and interference from grid lines. Therefore, it is necessary to translate the graphs of the well logging parameters into XY coordinates, where X denotes parameter values ​​and Y values ​​depth.

Bitmap records are image files created by scanning paper records. Saving good log data as depth-calibrated bitmaps is a cost-effective alternative to digital formats to preserve this essential information. Although bitmap well records are frequently deleted after vector conversion, they may hold a universal, computer-readable design key for old hardcopy data. This historical data is held on many media and provides information for various purposes, including environmental protection, water management, global change studies, primary and applied research, and resource exploitation and development.

Geologists and reservoir engineers, for example, return and analyze raster records manually or using software solutions that require a great deal of human input. Apart from wasting thousands of working hours, the current method is incorrect and time consuming. To digitize these raster registers and use them efficiently in conventional and unconventional analysis, one has to purchase an expensive digitizer, which is a manual and time-consuming task. There is also hidden technical debt because companies risk losing money on additional service and advisory fees. SCTR is the basis for Neuralog’s commercially available registration curve digitization software. However, this program frequently stops during curve tracking due to interference from the backend network.

Several unsupervised computer vision approaches have been created to digitize the log data embedded in the binary image. There are two ways to score well digitization: pixel-based methods and non-pixel-based methods. Both the mitigation procedure and the global curves trend approach are pixel-dependent methods. The thinning system reduces the line width to one pixel, leaving only the skeleton to represent its properties. The softening process is time consuming, loses line width information, and is prone to distortion and improper branching at intersections. The GCV approach is suitable for line processing but is not good for score line processing. Non-pixel-based methods are broadly categorized into two types: contour-based and contour-based.

The contour-based technique first involves extracting the contour of the image and locating the matching pairs of contours. The adjacency graph technique begins by encoding the graphs by encoding the run length, then analyzes the segments and generates various adjacency graph structures, such as the line affinity graph and the block-adjacent graph, using the SCTR methodology and the LAG data structure. Yang enhanced the SCTR approach and proposed the PCTR method. Based on BAG, Yuan and Yang devised a plan to delete grid lines and restore strokes in Chinese handwriting. However, such approaches are difficult to apply when dealing with complex scenarios in well-registered parameter graphs, especially node analysis. Yuan and Yang used morphological image processing and pixel statistics to isolate curves and gridlines.

The remaining grid lines and noise points are then scanned based on the modest size of the relevant components. However, all current technologies require manual intervention, which is not ideal, especially when paper records are more than 10MB wide. This search is from Deepkapha.ai IIT Kharagpur researchers suggest VeerNet, a new switch-based deep learning network that uses self-attention processes to learn distinct curves from a single pass. The simple design of the switches allows for multiple processing methods using comparable processing blocks.

Switches are well adapted for large-scale deep neural networks and large data sets. These advantages have led to impressive advances in different vision tasks using Transformer Khan et al. They train their model on synthetic and actual raster records. VeerNet operates with a resolution of 0.94, 0.48, and 0.39 in a multi-partition model of three-curve path digitization. VeerNet trained on accurate data has an accuracy of 0.6 for three curves.

Finally, their proposed approach to digitizing point-well log images is simpler and requires less manual involvement than previous solutions. The solution is also fast and scalable. They have developed VeerNet, a deep learning network that can effectively identify curves of good log and achieve a classification accuracy of more than 35%. The model can accurately distinguish the curves of a good log from the background grid, which improves on previous techniques.

This Article is written as a research summary article by Marktechpost Staff based on the research paper 'Digitization of Raster Logs: A Deep Learning Approach'. All Credit For This Research Goes To Researchers on This Project. Check out the paper.

Please Don't Forget To Join Our ML Subreddit