Hyperspectral technique combined with deep learning algorithm for detection of compound heavy metals in lettuce

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Heavy metal contamination in crops poses risks to human health.
• Accurate and efficient detection methods are crucial for food safety.
• Hyperspectral imaging offers a non-destructive approach to analyze crop quality.

Purpose of the Study:

• To develop a deep learning model for detecting compound heavy metals in lettuce.
• To extract deep features from hyperspectral images for heavy metal prediction.
• To evaluate the model's performance in predicting cadmium (Cd) and lead (Pb) content.

Main Methods:

• Visible-near infrared (400.68-1001.61 nm) hyperspectral imaging of lettuce.
• Wavelet Transform (WT) for multi-scale decomposition of spectral data.
• Stack Convolution Autoencoder (SCAE) for deep feature extraction.
• Support Vector Machine Regression (SVR) for content prediction.

Main Results:

• The WT-SCAE method effectively extracted deep features for heavy metal detection.
• High prediction accuracy achieved for Cd (R_p²=0.9319, RMSEP=0.04988 mg/kg, RPD=3.187) and Pb (R_p²=0.9418, RMSEP=0.04123 mg/kg, RPD=3.214).
• The model demonstrated robust performance in predicting heavy metal concentrations.

Conclusions:

• The combination of hyperspectral techniques and deep learning (WT-SCAE) shows significant potential for detecting compound heavy metals in lettuce.
• This method provides a reliable tool for non-destructive monitoring of heavy metal contamination in agricultural products.
• Further research can explore broader applications in food safety and quality assessment.