Tap Water Fingerprinting Using a Convolutional Neural Network Built from Images of Coffee-Ring Effect
Environmental Forensics
Oral Presentation
Prepared by X. Li1, R. Lahr2
1 - Michigan State University, 1449 Woodlot drive, Engineering Research complex A27, East lansing, MI, 48823, United States
2 - 1449 Engineering Research Complex, Room A127, East lansing, MI, 48823, United States
Contact Information: xiaoyanli629@gmail.com; 412-418-5816
ABSTRACT
Monitoring water quality efficiently and effectively has been a challenging but critical problem influencing people’s health. In this research, we proposed a low-cost and fast water monitoring technique using convolutional neural network (CNN) to analyze tap water fingerprints produced through the coffee ring effect. Twelve synthetic water samples were prepared, and residue pattern images were collected after drying droplets of each tap water on aluminum slides using the coffee ring effect. Principal component analysis (PCA) was conducted on both measurements of particles in the residue images and on the images themselves. PCA could not classify all the water components effectively, so a CNN model was built to classify images. Thirty tap water samples were collected from different cities in Michigan. Water chemistry data was measured, including Na+, Ca2+, Cu2+, Fe3+, PO43- and other elements by ICP-OES and IC. Using cluster analysis on the water chemistry data, these thirty samples were classified into six groups. Tap water sample droplets were dried on aluminum slides to harness the coffee ring effect and the residue patterns were collected with ten replicates. Six replicates were randomly selected from each water sample for the CNN model training dataset and the remaining replicates were used for the testing dataset. Ten individual convolutional neural network(CNN) models were trained and tested for their ability to classify residue images into groups with similar water chemistry. The classification accuracies of these ten models were all above 70%.The result suggests that although PCA is insufficient for classifying the images, a CNN model can effectively recognize coffee ring patterns of tap water fingerprints to assign them to water chemistry.
Environmental Forensics
Oral Presentation
Prepared by X. Li1, R. Lahr2
1 - Michigan State University, 1449 Woodlot drive, Engineering Research complex A27, East lansing, MI, 48823, United States
2 - 1449 Engineering Research Complex, Room A127, East lansing, MI, 48823, United States
Contact Information: xiaoyanli629@gmail.com; 412-418-5816
ABSTRACT
Monitoring water quality efficiently and effectively has been a challenging but critical problem influencing people’s health. In this research, we proposed a low-cost and fast water monitoring technique using convolutional neural network (CNN) to analyze tap water fingerprints produced through the coffee ring effect. Twelve synthetic water samples were prepared, and residue pattern images were collected after drying droplets of each tap water on aluminum slides using the coffee ring effect. Principal component analysis (PCA) was conducted on both measurements of particles in the residue images and on the images themselves. PCA could not classify all the water components effectively, so a CNN model was built to classify images. Thirty tap water samples were collected from different cities in Michigan. Water chemistry data was measured, including Na+, Ca2+, Cu2+, Fe3+, PO43- and other elements by ICP-OES and IC. Using cluster analysis on the water chemistry data, these thirty samples were classified into six groups. Tap water sample droplets were dried on aluminum slides to harness the coffee ring effect and the residue patterns were collected with ten replicates. Six replicates were randomly selected from each water sample for the CNN model training dataset and the remaining replicates were used for the testing dataset. Ten individual convolutional neural network(CNN) models were trained and tested for their ability to classify residue images into groups with similar water chemistry. The classification accuracies of these ten models were all above 70%.The result suggests that although PCA is insufficient for classifying the images, a CNN model can effectively recognize coffee ring patterns of tap water fingerprints to assign them to water chemistry.
