Back to results
US 12,339,224 B2Granted
Method and apparatus for determination of textile fiber composition
AssigneeValvan NV
InventorsFrank Vandeputte, Maurits Vandeputte, Peter Vandeputte
Life of this patent
Estimating the term…
Abstract
The current invention concerns determining a fiber composition of a textile sample. The sample is irradiated with near infrared (NIR) light. Reflected NIR light from the sample is captured. A vector of spectral values is determined based on the captured light. The vector is input to a deep neural network (DNN). The DNN comprises a sequence of layers of nodes, in particular an input layer, at least two intermediate layers, and an output layer. The nodes of successive layers of the sequence are interconnected via weighted edges. The DNN outputs for each of a plurality of fiber material types a numerical relative composition amount.
Claims20
A method for determining a fiber composition of a textile sample, comprising the steps of:
- moving the textile sample relative to a test point using an apparatus comprising a means for transporting the textile sample;
- irradiating the textile sample with near infrared light at the test point while the textile sample is moving relative to the test point via the transporting means;
- capturing reflected near infrared light from the textile sample at the test point while the textile sample is moving relative to the test point via the means for transporting the textile sample;
- determining a vector of spectral values based on the captured light;
- inputting the vector to an artificial neural network comprising nodes and edges, wherein each edge comprises a weight;
- and
- obtaining output from the artificial neural network,
- wherein the artificial neural network is a deep neural network comprising a sequence of layers of nodes, wherein the sequence comprises an input layer, at least two intermediate layers and an output layer, wherein nodes of successive layers of the sequence are interconnected via weighted edges, wherein the deep neural network is a feedforward neural network which is configured for data flow from the input layer to the output layer without looping back,
- wherein the output from the artificial neural network comprises for each of a plurality of fiber material types a numerical relative composition amount,
- wherein the method comprises the additional steps of:
- obtaining a vector of spectral values for each training textile sample of a multitude of training textile samples, wherein each of the plurality of fiber material types is present in at least one of the multitude of training textile samples, wherein at least several training textile samples of the multitude comprise a training composition of two or more fiber material types, via:
- irradiating the training textile sample with near infrared light,
- capturing reflected near infrared light from the training textile sample, and
- determining the vector based on the captured light;
- training the deep neural network with the vector of spectral values and a training composition of each training textile sample of the multitude,
- wherein, the multitude of training textile samples comprises a set of training textile samples with an identical fiber composition and with different humidity level, different coating, different color and/or different texture over the training textile samples of the set.
13 more elements
An apparatus for determining a fiber composition of a textile sample, comprising:
- a means for transporting the textile sample, the means for transporting the textile sample capable of moving the textile sample relative to a test point;
- a light source configured for irradiating the textile sample with near infrared light at the test point while the textile sample is moving relative to the test point via the means for transporting the textile sample;
- a capturing element configured for capturing reflected near infrared light from the textile sample at the test point while the textile sample is moving relative to the test point via the means for transporting the textile sample;
- a spectrometer configured for determining a vector of spectral values based on the captured light;
- a computer system, comprising at least one processor, configured for:
- obtaining the vector from the spectrometer,
- inputting the vector to an artificial neural network comprising nodes and edges, wherein each edge comprises a weight, and
- obtaining output from the artificial neural network,
- wherein the artificial neural network is a deep neural network comprising a sequence of layers of nodes, wherein the sequence comprises an input layer, a least two intermediate layers and an output layer, wherein nodes of successive layers of the sequence are interconnected via weighted edges, wherein the deep neural network is a feedforward neural network which is configured for data flow from the input layer to the output layer without looping back, wherein the output from the artificial neural network comprises for each of a plurality of fiber material types a numerical relative composition amount,
- preferably wherein the apparatus comprises a sensor head and a fiber optic cable in between the sensor head and the spectrometer, preferably wherein the sensor head comprises the light source and the capturing element, preferably wherein the spectrometer comprises a dispersing element and a detector,
- preferably wherein the apparatus comprises a conveyor belt for transporting the textile sample.
8 more elements
Dependencies are inferred from the claim text.