Classifying cadmium contaminated leafy vegetables using hyperspectral imaging and machine learning.

Area of Science:

• Agricultural Science
• Plant Physiology
• Analytical Chemistry

Background:

• Cadmium (Cd) is a toxic heavy metal that accumulates in edible plants, posing risks to human health.
• Conventional Cd detection methods are slow, costly, and environmentally burdensome, hindering efforts to mitigate plant uptake.
• Rapid and accurate methods are needed to assess Cd levels in crops.

Purpose of the Study:

• To evaluate the efficacy of hyperspectral imaging (HSI) combined with machine learning (ML) for predicting Cd concentrations in kale and basil.
• To identify key spectral bands indicative of Cd accumulation in plants.
• To compare the performance of different ML models for Cd classification.

Main Methods:

• Experiments were conducted in a controlled phenotyping facility using kale and basil.
• Visible/near-infrared (VNIR) hyperspectral images were acquired at harvest.
• Reflectance spectra were processed, and feature selection (ReliefF, PCA) was applied to train artificial neural network (ANN), ensemble learning (EL), and support vector machine (SVM) models.
• Plants were classified based on Cd concentration relative to a 0.2 mg kg^-1 safety threshold.

Main Results:

• Specific wavelengths (519-574 nm and 692-732 nm) were identified as crucial for Cd detection, correlating with changes in chlorophyll and leaf structure.
• All tested ML models successfully classified plants based on Cd levels.
• The artificial neural network (ANN) model achieved the highest F1 score for predicting Cd concentrations using all spectral wavelengths.

Conclusions:

• Hyperspectral imaging and machine learning provide a promising, rapid, and precise approach for diagnosing Cd contamination in leafy green vegetables.
• This non-destructive technique can accelerate the development of strategies to reduce Cd uptake in crops.
• Further research is necessary to adapt HSI-ML methods for real-world field conditions.