Breaking boundaries: harnessing unrelated image data for robust risky event classification with scarce state of polarization data
Khouloud Abdelli, Matteo Lonardi, J. Gripp, Samuel L. I. Olsson, Fabien Boitier, Patricia Layec
Abstract
We present an innovative transfer learning method for classifying risky events with scarce state of polarization (SOP) data, utilizing a deep convolutional neural network pre-trained on unrelated images. Achieving a 96.3% accuracy on just 400 samples, this approach offers a robust solution for data-limited scenarios.
Topics & Concepts
Convolutional neural networkComputer scienceArtificial intelligenceTransfer of learningPolarization (electrochemistry)Pattern recognition (psychology)Robustness (evolution)Deep learningData modelingMachine learningData miningDatabasePhysical chemistryChemistryGeneBiochemistryMachine Learning in Materials ScienceMetabolomics and Mass Spectrometry StudiesCognitive Science and Mapping