Séminaire du 10/12/2018, 11h00
lundi 10 décembre 2018, 11h00
Vision based evaluation of microbiological contamination levels using high-resolution images
Conférencier : Aurélien Launay
Microbiological contamination of a biological sample can be detected and measured through the detection and analysis of the growth of microorganism colonies observed in Petri dishes. The gold standard, based on the analysis of the Petri dishes through naked eye observation, or, more recently, using automated colony counters, requires waiting until the end of the incubation. To overcome this limitation, a new approach, relying on image captures during incubation and their analysis based on machine learning algorithms, is proposed. The currently available method used for automated counting further exhibits limits in terms of accuracy, due to difficulties related to image capture during incubation. In order to take account of these shortcomings, the proposed method makes use of convolutional neural networks, specifically trained to differentiate a microbiological colony from the background in a single image. This approach already demonstrates improvements in performance, in term of recall and precision, despite remaining residual false positives and false negatives, and its value when applied to each single image acquired during the incubation. Possible methods, using the whole set of images, to improve performances will also be presented.