Airborne systems capture the open mine terrain and geography as aerial images and analyse them manually to provide valuable insights. This process is effort-intensive and error-prone with a long lead time.

There is a need for acquiring spatial data remotely and more frequently and integrate the data with systems that automate the ingestion, analysis and publication processes. This approach automates the entire value chain and ensures the safety of the workforce.

The client required a partner with strong consulting and execution expertise in deep learning and computer vision and who could help achieve the defined business goals.

Key Challenges

  • High frequency of data capture, complex aerial imageries and large data sizes
  • Produce road feature data sets such as road centerlines, road edges and boundary edges of active mining areas using spatial elevation and imagery data and GPS snippet data for coal mine site
  • Training computer vision and deep learning algorithms to extract features at different altitudes

Ready to experience?

TALK TO EXPERTS
Line

The Solution

Deep learning algorithms on the objects with different altitudes

  • Deep learning solution to demarcate road centerlines, road edges and boundary edges of active mining areas using elevation and imagery data and GPS snippet data for coal and iron ore mining sites.
  • Techniques included identifying and segregating features based on different depth or elevations and stitching together multiple images to identify objects

Deep learning algorithms for aerial imagery

  • Demarcate road centerlines, road edges and boundary edges of active mining areas
  • Automating the entire process using advanced deep learning techniques
Line

Benefits

   

Extraction of the following datasets which typically require considerable manual effort help as enablers for analysis of a range of safety-critical controls

Road centrelines

Road centrelines

Road edges

Road edges

Pit crest and toe (active mining) area perimeter edges

Pit crest and toe (active mining) area perimeter edges

This complete process was automated through image analytic techniques which helped reduce the execution time by 75% for a typical 1000 km radius area.