Colorimetric Sensor Reading and Illumination Correction via Multi-Task Deep-Learning.

Area of Science:

• Nanotechnology
• Computational Science
• Sensor Technology

Background:

• Colorimetric sensors offer accessible nanotechnology for measuring substance properties via color changes.
• Current interpretation methods (visual or controlled camera setups) limit real-world applications due to altered light conditions and lower resolutions.
• There is a need for robust methods to interpret colorimetric sensor data accurately in non-ideal environments.

Purpose of the Study:

• To develop and evaluate image processing and deep learning (DL) methods for accurate colorimetric sensor reading under non-uniform illumination.
• To compare the performance of independent versus joint (multi-task) DL models for denoising and regression tasks.
• To improve the accessibility and accuracy of colorimetric sensors for end-user applications.

Main Methods:

• Collected video recordings of colorimetric sensors measuring temperature to create a reference dataset.
• Augmented sensor images with non-uniform color alterations to simulate real-world conditions.
• Implemented and evaluated various DL architectures, including a multi-task model that disentangles luminance, chrominance, and noise.

Main Results:

• The best-performing DL architecture achieved a mean squared error (MSE) of 0.811±0.074[°C] and r²=0.930±0.007.
• This multi-task model demonstrated a significant improvement of 1.26[°C] MSE compared to independent denoising and regression tasks.
• The model accurately predicted sensor readings despite strong color perturbations and non-uniform illumination.

Conclusions:

• The proposed DL methodology enhances the accuracy of colorimetric sensor readings in altered illumination conditions.
• This approach increases the potential for large-scale accessibility of colorimetric sensors in point-of-care diagnostics and continuous health monitoring.
• The developed methods overcome limitations of visual interpretation and controlled camera setups, broadening sensor applicability.