Automatic classification of UVES calibration images

General Goals

• Develop a method to supplement the visual spot check of calibration images done during daytime by experts. The method should identify anomalous images.

• Knowledge transfer on ML techniques from MetricArts to ESO team to work

Technical goal

• Design a method to detect anomalous and defectuous images based on Deep Learning and clustering techniques

Resources

• Microsoft Azure environment
• Azure storage container
• CPU and GPU VM
• Jupyterlab
• Tensorflow framework
• Keras library
• Scikit-learn library
• Astropy library

Data sources

• UVES Calibration images
  • Bias Blue & Red
  • Flat Blue & Red
  • Arclamps Blue & Red
• UVES corrupted file list (Garching)
• Corrupted mock images (Nicolas Haddad)

Authors

• Roberto Munoz
• Roberto Gonzalez
• Joaquin Prieto
• Pedro Fluxa