Deep and Machine Learning Using SEM, FTIR, and Texture Analysis to Detect Polysaccharide in Raspberry Powders.

Area of Science:

• Food Science and Technology
• Analytical Chemistry
• Computational Science

Background:

• Accurate identification of raspberry powder composition is crucial for quality control and product development.
• Traditional methods may lack the precision required for differentiating subtle variations in polysaccharide content and type.

Purpose of the Study:

• To develop and evaluate artificial neural network (ANN) models for classifying raspberry powders based on polysaccharide profiles.
• To integrate Fourier transform infrared spectroscopy (FTIR) and Scanning Electron Microscopy (SEM) data with machine learning and deep learning approaches.

Main Methods:

• Fourier transform infrared spectroscopy (FTIR) was used to obtain absorbance spectra.
• Scanning Electron Microscopy (SEM) provided microparticle structural information.
• Multi-Layer Perceptron Networks (MLPNs) with texture descriptors and Convolution Neural Networks (CNNs) with bitmaps were developed and trained.

Main Results:

• The study successfully devised MLPN and CNN models for raspberry powder classification.
• High classification efficiency was achieved, demonstrating the models' ability to recognize microparticle homogeneity.
• The integration of spectral and textural data proved effective for identifying variations in polysaccharide content.

Conclusions:

• ANN models, particularly CNNs and MLPNs, offer a powerful and innovative approach for the precise identification of raspberry powders.
• Combining FTIR spectroscopy with image texture analysis via neural networks enhances the ability to differentiate powders based on polysaccharide composition.
• This methodology provides a foundation for advanced quality control in food ingredient analysis.