Machine Learning in detecting chemical compounds in measurements of Ion Mobility Spectrometry by Enigma Pattern

Some details

Overview: Enigma Pattern developed CNN models capable of detecting predetermined set of chemical compounds in IMS spectra of gaseous samples. The method is resilient to noise and changes in measurements originating from different ambient conditions. 

 

The secondary goal was to determine which parts of the measured

… spectrum is the most important to classification process. EP’s approach to dataset augmentation allowed for
efficient learning on limited training dataset, whose collection and labeling is especially expensive,
compared to other domains.

 

Results: Enigma Pattern has identified a subset of features comprising 25% of the original information. Model trained on such data retained 

over 96% of full model’s detection performance.

Technologies: Tensorflow, Keras