A novel spectroscopy-deep learning approach for aqueous multi-heavy metal detection.

Area of Science:

• Environmental Science
• Analytical Chemistry
• Artificial Intelligence

Background:

• Heavy metal contamination in water poses significant environmental risks.
• Traditional detection methods are often time-consuming and expensive.
• Artificial intelligence (AI), including machine learning (ML) and deep learning (DL), shows promise for analytical chemistry but requires substantial spectral data.

Purpose of the Study:

• To develop an efficient and rapid method for detecting heavy metal concentrations in mixed water samples.
• To overcome the data acquisition limitations of traditional methods for AI model training.
• To utilize deep convolutional neural networks for simultaneous detection of multiple heavy metals.

Main Methods:

• A new digital spectral imaging system was developed to rapidly collect spectral data.
• 3000 digital spectra from mixed heavy metal samples were acquired.
• End-to-end deep convolutional neural network regression models (ResNet-50, Inception V1, SqueezeNet V1.1) were trained to predict heavy metal concentrations.

Main Results:

• The ResNet-50 model demonstrated high accuracy in simultaneously detecting arsenic, chromium, and copper.
• A linear fitting coefficient exceeding 0.99 was achieved between true and predicted heavy metal values.
• The developed system provides rapid and efficient heavy metal detection in complex water matrices.

Conclusions:

• The study presents an effective AI-driven approach for rapid heavy metal detection in environmental water samples.
• This method overcomes the limitations of traditional techniques in terms of speed and data requirements for AI.
• The findings serve as a valuable reference for advancing intelligent analytical techniques in environmental monitoring.