Well-Log Raster Scale Detection via Grid Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting the scale type of well-log properties in raster images face challenges due to inaccuracies in optical character recognition and the need for time-consuming supervised machine-learning techniques, which struggle with diverse grid patterns and labeled training data generation.
Innovation Solution
An unsupervised machine-learning approach using edge detection, gridlines localization, noise filtering, and pattern recognition to automatically identify linear or logarithmic scale types in well-log images, employing edge pixel detection, gridline localization, and noise filtering to enhance pattern recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If optical character recognition is used to detect scale type from log header range values, then scale detection can be performed, but detection accuracy deteriorates due to noisy raster images and inaccurate location of start and end values
Solution Approach 1:
The patent extracts and removes the log header information from the analysis, focusing solely on the grid pattern in the plot segment. This eliminates the source of OCR errors and location inaccuracies, allowing scale detection to proceed based purely on visual grid pattern analysis without being affected by noisy text recognition
Solution Approach 2:
The patent introduces an intermediary computational approach that transforms the grid pattern detection into a mathematical classification problem. By computing gridline positions and analyzing their spacing relationships through algorithms rather than direct OCR, the system mediates between the raw image data and the final scale type determination, achieving higher accuracy
2Measurement precision
If supervised machine-learning techniques are used for grid pattern classification, then scale detection can be performed, but processing time increases due to time-consuming labeled training data generation
Solution Approach 1:
The patent implements a self-service system where the grid pattern classification is performed through direct computational analysis of the image data itself. The algorithm automatically computes gridline positions, calculates spacing relationships, and determines scale type without requiring any external training data, labels, or supervised learning processes
Solution Approach 2:
The patent replaces the mechanical process of supervised machine learning (data collection, labeling, training) with a direct mathematical computation system. By substituting the learning-based approach with algorithmic grid pattern analysis, the system eliminates the time-consuming training phase while maintaining classification capability
3Measurement precision
If supervised machine-learning techniques are used for grid pattern classification, then scale detection can be performed, but reliability deteriorates when encountering diverse grid patterns not included in training data
Solution Approach 1:
The patent creates a universal grid pattern analysis system that can handle diverse grid patterns through a single computational framework. By analyzing fundamental geometric properties (gridline positions, spacing relationships) rather than learning from specific examples, the system achieves multi-functionality across different grid types without requiring retraining or additional data
Solution Approach 2:
The patent focuses on analyzing invariant parameters of grid patterns (spacing relationships, geometric proportions) that remain consistent across different grid types. By detecting these fundamental parameter relationships rather than learning specific pattern instances, the system achieves high adaptability to diverse grid patterns not seen during any potential training phase
Data Source
Figure 1A~1D
Figure 2
Figure 3A
AI summary
A method for processing an image of a subterranean formation includes performing edge detection on the image to produce a first processed image. The method also includes performing gridlines localizer on the first processed image to produce a second processed image. The method also includes performing noise filtering on the second processed image to produce a third processed image. The method also includes performing pattern recognition on the third processed image to determine that the image that corresponds to the third processed image is linear scale type or logarithmic scale type.