论文标题
基于树脂识别代码的塑料杂志隔离计算机视觉技术的应用
Application of Computer Vision Techniques for Segregation of PlasticWaste based on Resin Identification Code
论文作者
论文摘要
本文提出了基于其树脂识别代码识别塑料废物的方法，以提供对消费后塑料废物的有效回收。我们建议（i）在训练系统训练时，建议（i）确定属于已知塑料废物类别的塑料废物的设计，培训和测试，并确定在系统训练时不属于任何已知类别的塑料废物类别的新塑料废物。对于第一种情况，我们建议使用暹罗和三重损失网络使用一声学习技术。我们提出的方法不需要任何增加来增加数据库的大小，并获得了99.74％的高精度。对于第二种情况，我们提出使用监督和无监督的降维技术，并获得95％的精度，以正确识别新的塑料废物。
This paper presents methods to identify the plastic waste based on its resin identification code to provide an efficient recycling of post-consumer plastic waste. We propose the design, training and testing of different machine learning techniques to (i) identify a plastic waste that belongs to the known categories of plastic waste when the system is trained and (ii) identify a new plastic waste that do not belong the any known categories of plastic waste while the system is trained. For the first case,we propose the use of one-shot learning techniques using Siamese and Triplet loss networks. Our proposed approach does not require any augmentation to increase the size of the database and achieved a high accuracy of 99.74%. For the second case, we propose the use of supervised and unsupervised dimensionality reduction techniques and achieved an accuracy of 95% to correctly identify a new plastic waste.
