US 11,865,740 B2
Systematic disposal, classification and dynamic procurement of recyclable resin
Swati Murthy, Bangalore (IN); and Rameshwar Gongireddy, Hyderabad (IN)
Assigned to Tata Consultancy Services Limited, Mumbai (IN)
Filed by Tata Consultancy Services Limited, Mumbai (IN)
Filed on Jun. 15, 2021, as Appl. No. 17/348,263.
Claims priority of application No. 202021025088 (IN), filed on Jun. 15, 2020.
Prior Publication US 2022/0019842 A1, Jan. 20, 2022
Int. Cl. B29B 17/02 (2006.01); G06V 10/44 (2022.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01); G06F 18/2413 (2023.01); G06V 10/82 (2022.01)
CPC B29B 17/02 (2013.01) [G06F 18/2413 (2023.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06V 10/454 (2022.01); G06V 10/82 (2022.01)] 9 Claims
OG exemplary drawing
 
1. A processor implemented method, comprising:
obtaining, via one or more hardware processors, plurality of input images of a plastic item using an image sensor, a location of the plastic item using a location sensor, and weight of the plastic item using a weight sensor;
predicting, by a trained Convolutional Neural Network (CNN) model, a type of resin associated with the plastic item, based on the plurality of input images, the location and the weight of the plastic item, via the one or more hardware processors, wherein training the CNN model comprises:
receiving a training data at the CNN model, the training data comprising a plurality of training images captured through various orientations of a plurality of plastic items, and weight and location data of the plurality of plastic items,
extracting, by using a plurality of filters, a plurality of features from the training data by processing an image data associated with the plurality of input images to obtain a plurality of physical attributes, and processing non-image data comprising the weight and location of the plurality of plastic items, wherein the image data is processed by a convolution component of the CNN model and the non-image data is processed by a feedforward component of the CNN model, and wherein one or more localized areas of the plurality of training images are computed through Histogram of Oriented Gradients;
mapping the plurality of features with a set of resin identification codes based on the plurality of physical attributes to identify a label for the plastic item, wherein the label comprises a type of resin of the plastic item; and
associating the type of resin associated with the plastic items of the training data based on the plurality of features; and
sharing, via the one or more hardware processors, the type of resin, weight and location of the plastic item with a sever.