Master Thesis - Deep Learning from Teacher Sensors

Deep Learning opens up new opportunities to solve machine vision applications. It also opens up for simplification of previously complicated tasks. This may relate to the effort needed to integrate and configure the solution, but also that solutions can now rely on less complex and more affordable sensor technologies. What previously required a hyperspectral camera might be solvable with an RGB camera. A 3D problem might now be possible to handle using a monochrome 2D camera, or maybe even just from other production line sensors that anyway need to be there.

Deep Learning may thus make previously too complex or expensive problems solvable, profitable or simply just affordable. This can further optimize production and reduce waste in such diverse fields as factory production, agriculture, logistics or process automation.

This thesis project aims to investigate how such downscaling could work in practice. By collecting data using a high-end sensor or algorithm in parallel with a simpler one, we hope to be able to show how the student sensor can learn from its teacher sensor how to interpret what it sees.

Prerequisites

The applicant should have experience from these fields.

• Image Processing
• Machine Learning
• Python scripting or similar
• TensorFlow, PyTorch or similar framework

Contact

For more information, contact R&D Manager Andreas Wrangsjö andreas.wrangsjo@sick.se

We look forward to your application before  the 18th of October!