Transfer learning method for plastic pollution evaluation in soil using NIR sensor.

Area of Science:

• Environmental Science
• Analytical Chemistry
• Machine Learning

Background:

• Plastic debris poses a significant environmental threat, necessitating effective soil monitoring.
• Accurate assessment of plastic pollution levels in diverse soil environments is crucial.

Purpose of the Study:

• To develop a transfer learning model for evaluating soil plastic pollution across different regions using an ultra-portable NIR sensor.
• To compare the efficiency and accuracy of Manifold Embedded Distribution Alignment (MEDA) and Transfer Component Analysis (TCA) with conventional Support Vector Machine (SVM) models.

Main Methods:

• Utilized an ultra-portable Near-Infrared (NIR) sensor for soil sample analysis.
• Applied transfer learning algorithms, specifically Manifold Embedded Distribution Alignment (MEDA) and Transfer Component Analysis (TCA).
• Compared transfer learning models against conventional Support Vector Machine (SVM) models for classification accuracy and running time.

Main Results:

• MEDA models achieved high average classification accuracy (97.78% in source, 79.52% in target regions).
• MEDA models demonstrated significantly faster processing times (0.70s) compared to TCA (21.90s) and SVM (41.38s).
• MEDA outperformed TCA and conventional SVM in both accuracy and efficiency for cross-region soil plastic detection.

Conclusions:

• Transfer learning, particularly the MEDA method, offers an efficient and accurate approach for detecting soil plastic pollution.
• The combination of an ultra-portable NIR sensor and MEDA provides a promising low-cost solution for field-based environmental monitoring.