Identifying 3D objects
AssigneesFundació Institut De CièNcies FotòNiques, Institució Catalana De Recerca I Estudis AvançAts
InventorsValerio Pruneri, Mehmet Alican Noyan, Anatoly Grudinin
Life of this patent
Estimating the term…
Abstract
Identifying 3D objects A method for identification of 3D objects comprises illuminating at least part of a 3D object with electromagnetic radiation, spectroscopically obtaining spectral data for one or more regions of the 3D object, and generating, at a data processing apparatus, an identification result for the 3D object using a trained machine learning model. The trained machine learning model processes the obtained spectral data for the one or more regions to generate one or more model outputs from which the identification result is derived.
Claims8
An apparatus, comprising:
- a spectroscopy arrangement to obtain a plurality of spectral data sets for a plurality of respective regions of a 3D object, the spectroscopy arrangement including a source of electromagnetic radiation configured for illuminating at least part of the 3D object;
- one or more processors;
- and
- one or more computer-readable media storing:
- a trained machine learning model configured to process the obtained spectral data sets to generate a plurality of respective model outputs from which an identification result is derived, each model output comprising a latent space position for a respective spectral data set, the trained machine learning model comprising a neural network model which has been trained to minimise a latent space distance between measurements for the same object and to maximise a latent space distance between measurements for different objects, and
- computer readable instructions, which when executed by the one or more processors, generate the identification result using the trained machine learning model, wherein generating the identification result comprises, for each of the plurality of obtained spectral data sets, comparing the latent space position for the obtained spectral data set with latent space positions for a plurality of stored spectral data sets for a plurality of regions of a reference 3D object, wherein the identification result thereby quantifies an extent to which the 3D object is related to the reference 3D object.
A method, comprising:
- illuminating at least part of a 3D object with electromagnetic radiation;
- obtaining a plurality of spectral data sets for a plurality of respective regions of the 3D object, using a spectroscopy technique, and
- generating, at a data processing apparatus, an identification result for the 3D object using a trained machine learning model, comprising processing the obtained spectral data sets using the trained machine learning model to generate a plurality of respective model outputs from which the identification result is derived, each model output comprising a latent space position for a respective spectral data set, the trained machine learning model comprising a neural network which has been trained to minimise a latent space distance between measurements for the same object and to maximise a latent space distance between measurements for different objects,
- wherein generating the identification result comprises, for each of the plurality of obtained spectral data sets, comparing the latent space position for the obtained spectral data set with latent space positions for a plurality of stored data sets for a plurality of regions of a reference 3D object, wherein the identification result thereby quantifies an extent to which the 3D object is related to the reference 3D object.
Dependencies are inferred from the claim text.