Dresden 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 2: AKPIK Posters
AKPIK 2.2: Poster
Montag, 16. März 2020, 18:45–19:30, P2/1OG
Real-Time Localization and Classification for Digital Microscopy using Single-Shot Convolutional Neural Networks — Martin Fränzl and •Frank Cichos — Molecular Nanophotonics Group, Peter Debye Institute for Soft Matter Physics, Universität Leipzig, Germany
Common single particle tracking approaches work fast and reliable for images, where all objects have a sufficiently high signal to noise. Whenever the contrast in the images varies considerably over the particles and the signal-to-noise is low most simple techniques fail. We present a single shot neural network system based on the YOLO architecture for the classification and tracking of particles in optical microscopy data. The network is implemented in Python/Keras using the TensorFlow backend. The trained model is then exported to a GPU supported TensorFlow C library for real-time inference readily integrable in other programming languages such as MATLAB and LabVIEW. The network is able to reliably classify and localize several hundred objects in images of 416 × 416 pixels even at low signal to noise (SNR ∼ 1). As compared to previous work, our system is fast to allow for a real-time detection at 50 frames per second independent of the number of particles contained.