Model development for Image Object Classification and OCR analysis for mining industry
Key Challenges
Key Results
Overview
Model development for Image Object Classification and OCR analysis for mining industry
Minalyze AB was formed in Gothenburg, Sweden, in 2009. Minalyze is the world's leading manufacturer of XRF core scanning devices and software for geological data display. The Minalyzer CS was introduced in 2014 and is capable of continuous XRF scanning of drill cores in core trays. The Minalyzer has scanned over 100 000 meters of cores, providing clients with important and usable data. Minalogger, a software for visualizing geological data, was also developed by Minalyze AB.
Challenges
Minalyze provides the mining and resources industries with a technology service offering to improve how data is measured from drillhole or well data. This includes recording sensor data from X-rays (known as X-ray fluorescence), laser telemetry for point cloud data and high-resolution image data. The Minalyze team wanted to develop custom Machine Learning Models for Image Object Classification and OCR analysis of core images with the ability to provide a YAML or JSON request to an API endpoint. They also wanted to have a repeatable solution as Infrastructure As Code where complex machine learning pipelines can be deployed easily to increase speed of their data output and reduce errors.
Solution
- The goal was to build and train Machine Learning Models for Image Object Classification and OCR analysis. The Ankercloud team developed two Machine Learning models to meet the outcome using AWS Sagemaker.
- The primary Model was developed mainly for pre-processing, object recognition & classification and creation of bounding boxes. The secondary Model was developed to perform Optical Character Recognition on the core blocks.
- The developed machine learning model detects wooden or plastic blocks known as core blocks within the images of core trays and calculates a bounding box. It also identifies and classifies any written marking on the core rock itself and calculates a bounding box. It then performs OCR on the core blocks to record what text values are written. Lastly, the model creates a JSON object with the bounding box information (box center, dimensions) and retrieved text so that it can be related back to the images.
- We have used Amazon S3 to store data for training. Amazon SageMaker Notebooks and Amazon Sagemaker Training Instances to train the model and save the latest model to Amazon S3. It is accessed from the Amazon SageMaker Inference endpoints. We also created a ML-Ops pipeline which includes steps for pre-processing, training, evaluation, conditional evaluation, and model registration using Amazon Sagemaker tools.
Business Outcome
- Ready to use and preconfigured Amazon Sagemaker process for Image object classification and OCR analysis Models along with ML- Ops pipeline.
- Increase in speed and accuracy of object classification and OCR leads to increased operational efficiency.
- A working Infrastructure on AWS that can scale accordingly to the growth and is easily expandable with different AWS services.
- Minalyze intends to use this project as a template for future machine learning engineering projects.
