Correlation awareness evolutionary sparse hybrid spectral band selection algorithm to detect aflatoxin B1...

Area of Science:

• Agricultural Science
• Food Science
• Analytical Chemistry

Background:

• Aflatoxin B1 contamination in almonds poses significant health risks.
• Hyperspectral imaging offers rapid, non-destructive detection but faces challenges with high dimensionality and complex data analysis.
• Accurate and efficient detection methods are crucial for ensuring food safety in industrial settings.

Purpose of the Study:

• To develop a novel hybrid spectral band selection algorithm for classifying aflatoxin B1 in almonds.
• To simplify hyperspectral data analysis for industrial applications.
• To improve the efficiency and accuracy of aflatoxin B1 detection in food products.

Main Methods:

• A two-step hybrid spectral band selection algorithm was proposed.
• Significant spectra were identified using tree-based boosting ensembles and multilayer perceptron networks.
• Correlation-aware sparse spectral band selection optimized the identified spectra.

Main Results:

• The algorithm achieved comparable classification accuracy to full spectra models using only 4 to 10 selected spectra.
• Evaluation on three hyperspectral image datasets demonstrated the algorithm's effectiveness.
• The method proved suitable for industrial applications requiring rapid and non-destructive testing.

Conclusions:

• The developed hybrid spectral band selection algorithm effectively reduces hyperspectral data complexity for aflatoxin B1 classification in almonds.
• The selected spectral bands provide a robust and efficient method for industrial food safety applications.
• This approach enhances the practical utility of hyperspectral imaging in detecting food contaminants.