Low-Cost Recognition of Plastic Waste Using Deep Learning and a Multi-Spectral Near-Infrared Sensor.

Area of Science:

• Materials Science
• Computer Science
• Environmental Science

Background:

• Accurate plastic identification is crucial for effective recycling and waste management.
• Current methods can be expensive or inaccessible for widespread use.
• Developing low-cost, portable solutions is essential for advancing sustainable practices.

Purpose of the Study:

• To present a novel approach for plastic recognition using a low-cost spectroscopy sensor and machine learning.
• To validate the effectiveness of various machine learning algorithms for classifying common household plastics.
• To demonstrate the potential of this technology for affordable and portable plastic identification.

Main Methods:

• Utilized a multi-spectral sensor measuring 18 wavelengths (visible to near-infrared).
• Employed ten machine learning algorithms including Convolutional Neural Networks (CNNs) and Multi-Layer Perceptrons (MLPs).
• Collected and analyzed data from six plastic types: PET, HDPE, PVC, LDPE, PP, and PS household waste.

Main Results:

• CNNs achieved a mean accuracy of 72.50% and MLPs achieved 70.25% in plastic recognition.
• Highest accuracy was 83.5% for Polystyrene (PS), lowest was 66% for Polyethylene terephthalate (PET).
• The developed pipeline demonstrated effective plastic recognition capabilities.

Conclusions:

• Low-cost near-infrared spectroscopy combined with machine learning offers an effective solution for plastic recognition.
• This approach is affordable, portable, and contributes to sustainable systems.
• Potential applications extend to agriculture, e-waste recycling, healthcare, and manufacturing.