Prof. Dr. Martin Regehly

Phone +49 3381 355-385
Fax +49 3381 355-199

regehlyth-brandenburg.de

Contact Us

Visual Sciences / Optical Systems Engineering
Department of Engineering

Brandenburg University
of Applied Sciences
Magdeburger Straße 50

14770 Brandenburg an der Havel


Administrative inquiries:
Baraa Asfari
Lab Manager
asfarith-brandenburg.de

You are here:ResearchProjects

Current Topics

Deep Convolutional Neural Networks for Image Classification and Segmentation

We are establishing an environment to explore the capabilities of convolutional deep neural networks for image classification and segmentation. We are using Keras, an open source neural network library for Python which runs on top of Tensorflow, both developed by Google. As a first example we are training a convolutional network for segmentation of retina images. Our goal is to extract  important features like the optic disk or the vessel system (see left image for an output of the network) which are fine markers for potential diseases of the eye or systemic pathological changes. Training of deep neural networks needs extensive computational power, depending on the architecture, depth and size of the network as well as amount of annotated training data. We are following a cost-effective approach using standard but yet powerful GPU graphics cards supporting CUDA.

 


Completed Projects

Scientific Superresolution Cameras

Scientific cameras exhibit larger pixel sizes to achieve high dynamic range measurements. This comes at the expense of lower resolution as the number of pixels fitting into the sensor is limited. Increasing the sensor size is often undesirable because of drastic cost increase and required large optical systems. An interesting approach is to acquire several low resolution frames whereas the sensor is sub-pixel shifted. Afterwards a reconstruction algorithm calculates a superresolution image from low res frames. In corporation with the Max-Born-Institute, this project dealed with the development of an integrated subpixel displacement unit based on a compact x,y piezo-stage, capable to work with deep peltier-cooled scientific image sensors. A superresolution camera including software for reconstruction had been developed and brought to market. The prototype was tested in the soft X-ray range, the gain in resolution has been determined to a factor of 1.3 for each axis, a value close to the theoretical limit of 1.5 for this technique. 
 

Wafer Scale XUV Imaging Detector

For demanding low light level imaging applications in the extreme ultraviolett range (XUV), a class of high performance science cameras had been developed. Due to its large area of approx. 60mm x 60mm, the CCD sensor is fabricated by e2V on a single wafer. The latter makes the camera extraordinary expensive as many wafers must be processed to obtain a sensor reaching the specs. Concurrently deep peltier cooling downto -90oC makes the thermal design challenging. The sensors features 4 video outputs, therefore 4 analog front-end need to be integrated and a digital back-end combining all 4 data streams. Due to the large pixel size, signal-to-noise ratio of the sensor is up to 175000:1 The camera can be used for diffraction imaging, spectroscopy and XUV microscopy. It is typically sold to large scale research facilities like free electron lasers or dedicated synchrotron beam lines.

In-vacuum Detectors for Imaging and Spectroscopy

Based on the ultra-low noise electronics platform developed previously, a series of in-vacuum cameras with various CCD sensor formats and technologies were designed and manufactured. The work undertaken included a redesign of circuit boards and internal components for low outgassing properties in addition to a liquid coolant system to dissipate the heat generated by the internal multistage peltier elements. Further activities included to establish a workflow for manufacturing such stainless steel cameras including bake-out procedures. These detectors are typically used for imaging and spectroscopy of UV, extreme UV and soft X-ray radiation. One main driving force is EUV lithography as an enabling technology for the fabrication of next-generation processors. Cameras are also actively used in fundamental research in large scale facilities like DESY Hamburg or BESSY.  
 

Cell and Wafer Characterisation System

A multivalent tool for characterisation of wafers along the production chain from as-cut silicon wafers to solar cells has been developed together with the Humboldt-University and introduced to the market. The system integrates electroluminescence (EL), photoluminescence (PL) and lock-in thermography (LIT) inspection capabilities for detailed quality and process control. It was the world first device employing high-power LEDs instead of lasers as an excitation source significantly reducing cost & complexity. For the compactness, versatility and innovative imaging technologies, the system has been conferred with the Berlin-Brandenburg innovation ward. The tool is used by cell manufacturing plants in asia as well as in many international research institutions.     
 

Outdoor Inspection System for Photovoltaics

Using an inverse process, solar modules emit near-infrared light when excited by an electric current. This in turn allows to detect a variety of possible defects within the modules which are typically invisible. The project involved the development of a portable inspection system, capable of imaging the faint electroluminescence from solar modules after dusk. It employs a sensitive NIR camera, a portable battery coupled to a DC-DC upconverter and a software system. Background subtraction and filter techniques remove the much brigher background signal of the sky, being the first commerical system of that kind. The award-winning system is used by researchers and solar companies in several countries.
 

Portfolio of High-Performance Scientific Cameras

A complete range of deep-cooled scientific cameras for UV, VIS and NIR imaging and spectroscopy had been developed. The cameras feature an ultra-low noise front-end for high performance CCD sensors from e2V and an FPGA based control and interface back-end. The development work undertaken included electronic circuit and pcb design, mechanics & housing, multistage peltier deep cooling, vacuum tight sealed chambers as well as driver and software development. Overall more than 20 different models have been brought to market and sold to customers in research and industry worldwide.

 



Electroluminescence Imaging System


 
The system utilizes the electroluminescence phenomenon to image micro-cracks and other defects of photovoltaic modules which are extremely difficult or impossible to detect visually. For this purpose an NIR sensitive scientific camera has been employed to detect the faint phonon assisted band-to-band recombination luminescence of silicon. A linear stage shifts the camera stepwise across the solar panel to increase the resolution. The system was the first commercially available system for panels on the world market. It had been sold around the globe for inspection and research of photovoltaics.
Early works:

High Resolution Panorama Scanner



 
The scanner is based on a color CCD line sensor with about 8000 pixels. Through precise rotational movement of the CCD by a linear motor with incremental encoder, the surrounding scene is captured. Images with up to 8000 x 50000 pixels were acquired, resulting in image sizes in excess of 1GB per picture. The scanner has been used for landscape and architecture high res imaging. Fast moving objects generate artefacts, inherent to the line scan technique. The project won the first price during the Jugend forscht regional and state competition in Berlin as well as the second price on the national level in Germany.