Image-Based Machine Learning Using Inkjet-Printed Chemicals: Mixing Ratio Prediction and Metal Ion Detection.

Abstract

Inkjet printing of π-conjugated organic compounds enabled rapid, low-cost generation of training images for the image-based machine learning (ML) prediction of mixing ratios. ML models with mean absolute errors below 4% were achieved within hours, even for dyes with subtle color differences. Changing the printing surface from filter paper to a polypropylene film extended the method to colorless compounds, including isomeric and macrocyclic systems. This approach also enabled spatial mapping of sub-microgram levels of Zn2+ ions using a weakly responsive colorimetric sensor, without the need for a spectrometer. This work demonstrates a simple, versatile strategy for integrating π-conjugated materials with ML in colorimetric sensing and mixture analysis.