Efficient, automatic, and optimized portable Raman-spectrum-based pesticide detection system.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Raman spectroscopy offers precise pesticide detection and chemical composition analysis.
• Portable spectrometers are valuable for field applications but face challenges with noise and signal offsets.
• Principal Component Analysis (PCA) is a common method for Raman spectrum identification, but its accuracy is limited by data imperfections.

Purpose of the Study:

• To develop an improved method for accurate pesticide identification using portable Raman spectroscopy.
• To overcome the limitations of traditional algorithms like PCA in handling noisy and offset spectral data.
• To optimize a Convolutional Neural Network (CNN) model for enhanced pesticide composition analysis.

Main Methods:

• Collected Raman spectra using a portable spectrometer.
• Preprocessed spectral data using a small-step, center-weighted moving-average method.
• Trained and optimized a CNN model using various algorithms, including cat swarm optimization, for prediction.
• Self-optimized data preprocessing and model architecture for improved data handling.

Main Results:

• The optimized CNN model demonstrated improved performance in pesticide identification compared to PCA.
• The developed model achieved an accuracy of 89.33% for identifying the composition of three different pesticides.
• Self-optimized data preprocessing and CNN architecture enhanced the model's ability to manage diverse spectral data.

Conclusions:

• A CNN model, optimized with cat swarm optimization and advanced data preprocessing, significantly improves pesticide identification accuracy from portable Raman spectroscopy.
• This approach offers a robust solution for field-based pesticide analysis, overcoming common spectral interferences.
• The study highlights the potential of deep learning in enhancing the reliability of spectroscopic methods for chemical analysis.