Machine-Learning-Based Color Sensing Using Wearable SENSIPATCH Spectrometer Module: An Experimental Study

Area of Science:

• Spectroscopy
• Wearable technology
• Machine learning

Background:

• Accurate color classification is vital in industries like textiles and biomedicine.
• Existing methods may lack portability or adaptability.
• Wearable sensors offer potential for on-site, real-time color analysis.

Purpose of the Study:

• To develop and evaluate a machine learning framework for spectral color classification using a wearable sensor.
• To assess the performance of different machine learning models for this task.
• To investigate the robustness of the system against variations in wearable sensor placement.

Main Methods:

• Utilized the spectrometer module of the SENSIPATCH wearable system, featuring multi-wavelength LEDs and photodiodes.
• Collected spectral data for 100 standardized PANTONE colors under controlled lighting.
• Applied data preprocessing techniques including baseline correction, bootstrapping, and Z-score normalization.
• Trained and evaluated five machine learning models: Random Forest, SVM, MLP, CNN, and LSTM.

Main Results:

• The Multilayer Perceptron (MLP) model exhibited the highest color classification performance.
• The MLP model demonstrated consistent accuracy across both firm and loose sensor contact scenarios.
• The results indicate robustness against variations in wearable placement.

Conclusions:

• Compact LED-based wearable spectroscopy is feasible for reliable color classification under controlled conditions.
• The developed MLP model shows promise for practical applications requiring accurate color sensing.
• Further research can extend this approach to variable lighting conditions.