Loading...

Evaluation of the Potential of Hyperspectral Imaging Combined with Chemometrics for Rapid Detection of Pesticides in Acetamiprid and Dimethoate Agricultural Products

Omrani Dizajyekan, Parya | 2026

2 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58897 (03)
  4. University: Sharif University of Technology
  5. Department: Chemistry
  6. Advisor(s): Parastar Shahri, Hadi; Garmroodi Asil, Ali
  7. Abstract:
  8. The widespread use of pesticides in modern agriculture, despite markedly improving crop productivity, has raised serious concerns regarding public health and food safety. Conventional approaches for detecting pesticide residues, including chromatography-based techniques, although highly accurate and sensitive, face substantial practical limitations due to being time-consuming, costly. Accordingly, the main objective of this research is to develop a rapid, and accurate method based on hyperspectral imaging in the 400-950 nm spectral range, combined with deep learning models built on Convolutional Neural Networks, to detect two widely used pesticides dimethoate and acetamiprid on the surface of green apples. In this study, five different concentration levels were prepared for each pesticide, and hyperspectral data were acquired from the corresponding samples. During preprocessing, the region of interest was extracted from the images, and the spectral range of 622-700 nm was selected as an information-rich region. To address limitations arising from the small number of physical samples, a spectroscopy-principled data augmentation strategy was applied, such that each of the 143 spectral bands was treated as an independent image during the model training process. Data analysis indicated that classical chemometrics methods, including Multivariate Curve Resolution, were unable to accurately separate pesticide signals due to severe interference from the apple’s biological matrix, particularly the dominant presence of chlorophyll. In contrast, CNN-based models, leveraging their capability to automatically extract spatio-spectral features, exhibited substantially more favorable performance for pesticide detection. To ensure result validity and prevent data leakage, a sample-based validation strategy was employed. The final results show that the optimized model after removing data associated with concentrations below the detection limit (less than 50 mg/L) and refining inconsistent samples achieved an approximate classification accuracy of 82%. These findings demonstrate that combining hyperspectral imaging with deep learning offers strong potential for developing efficient tools for rapid screening and quality monitoring of products in the food industry
  9. Keywords:
  10. Hyperspectral Imaging ; Deep Learning ; Convolutional Neural Network ; Apple ; Agricultural Products ; Pesticide ; Pesticides Residue

 Digital Object List

 Bookmark

No TOC