Deep learning-assisted self-volume-calibrating colorimetric sensor for chloride ion detection

Abstract

Chloride ions (Cl−) are widely present in nature, and accurate measurement of Cl− is of great significance for health management and water quality monitoring. Based on the principle of argentometry, where Ag+ react with Cl− in silver measurement to produce insoluble AgCl, we developed a colorimetric sensor composed of AgNO3, 2′,7′-dichlorofluorescein (DCFS) and starch, which is defined as ADS sensor. Cl− react with Ag+ in the ADS sensor to form positively charged AgCl colloids, which can adsorb negatively charged DCFS dyes and cause a color change in DCFS, thus providing sensing information about Cl− concentrations. To eliminate the error caused by the unfixed sample volume, a novel self-volume-calibrating strategy based on the unique shape of the sample holder was proposed. We further utilized deep learning (DL) algorithms to analyze data from 3,900 colorimetric images, enabling rapid determination of 1 × 10−5 to 7 × 10−2 mol/L Cl− with unfixed sample volumes. The Class Activation Mapping (CAM) algorithm was also used to visualize the decision-making mechanism of Convolutional Neural Network (CNN), verifying that the colorimetric regions and volume change regions of the sensor is the primary decision-making information for CNN. Compared with traditional method, this approach has the advantages of low cost, wide detection range, simple operation and rapid response. We envision that this method for Cl− detection can be used as an alternative and promising tool for the detection of a wider variety of analytes.