All Technologies Used
Motivation
The customer aimed to automate the process of reading data from graphs printed by equipment used to account for extracted oil and gas resources. The task involved processing graphical information, printed data (stickers with text, barcodes), and handwritten data (dates, numbers, and notes) from equipment operators.
Main Challenges
The equipment printed data on round discs, which were scanned and sent to the system for reading and processing. Azati developed an algorithm to unfold the round disk and convert the image into a rectangular format to trace and read the coordinates of the curves accurately.
The graphs contained multiple curves with different colors, sometimes overlapping with the background. Azati trained a neural network to accurately select and highlight curves of different colors, even when extraneous interference occurred, ensuring accurate reading of each curve.
The customer had multiple partners using different types of equipment, generating data in various formats. Azati created a neural network that identified the input data's format, categorized it, and routed it to the correct data processor for processing.
Handwritten data, such as dates and numbers, presented challenges due to variability in legibility. Azati used Google Tesseract along with a trained neural network to recognize handwritten data from multiple regions of the scanned images, overcoming issues caused by human factors like haste or poor handwriting.
Key Features
- High-Accuracy Barcode Processing: The AI-powered service achieved 90% accuracy in processing barcodes, ensuring that important information could be extracted accurately from scanned images.
- Accurate Line Recognition on Graphs: The service processed curves on graphs with an accuracy rate above 80%, allowing for the extraction of key data such as resource extraction metrics and other performance indicators.
- Handwriting Recognition: Handwritten data, such as dates and numbers, was processed with variable accuracy depending on input quality. The accuracy ranged from 30% to over 70%, with human factors influencing the result.
- Customizable Neural Network Processing: The neural network was trained to process various data formats based on customer-specific input, allowing the system to adapt to different types of equipment and data structures used by the customer's partners around the world.
Our Approach
Project Impact
Azati successfully developed and integrated AI-powered data processing services into the customer's infrastructure, significantly reducing manual data processing efforts.
The system provided highly accurate and automated recognition of barcodes, curves, and handwritten data, improving operational efficiency and accuracy for the oil and gas industry.