Maximizing the Accuracy of Continuous Quantification Measures Using Discrete PackTest Products with Deep Learning and...

Area of Science:

• Environmental chemistry
• Analytical chemistry
• Machine learning applications

Background:

• PackTest products offer a rapid method for estimating liquid sample chemical characteristics.
• Deep learning integration with colorimetric methods is an emerging area for quantitative analysis.

Purpose of the Study:

• To evaluate the accuracy and precision of PackTest products combined with deep learning for quantifying chemical oxygen demand (COD), ammonium, and phosphate ions.
• To explore the efficacy of a pseudocolor imaging method for water quality parameter determination.

Main Methods:

• PackTest products reacted with standard solutions, and resulting colors were scanner-read.
• Pseudocolor imaging generated multiple grayscale images from color data (RGB, CIELAB, CMYK).
• Deep learning models were trained on datasets derived from these grayscale images for quantification.

Main Results:

• Chemical oxygen demand quantification achieved high accuracy (normalized mean absolute error < 0.4%) and precision (coefficient of determination > 0.9996).
• Initial quantification of ammonium and phosphate ions yielded false positives.
• Multiple regression and optimized reference value estimation significantly improved accuracy for ammonium and phosphate ion quantification, respectively.

Conclusions:

• The combination of PackTest and deep learning is highly effective for accurate COD measurement.
• Methodological refinements are crucial for reliable quantification of ammonium and phosphate ions using this approach.
• Pseudocolor imaging with deep learning shows promise for water quality monitoring.