Related Experiment Video
Updated: Jul 4, 2025
06:28
Author Spotlight: Unraveling Plant Responses to Abiotic Stresses Using the PlantScreen Robotic Platform
Published on: June 7, 2024
1.8K
Research on detection of potato varieties based on spectral imaging analytical algorithm
You Li1, Zhaoqing Chen1, Fenyun Zhang1
1School of Automation, Hangzhou Dianzi University, Hanzhou, Zhejiang Province 310018, China.
Spectrochimica Acta. Part A, Molecular and Biomolecular Spectroscopy
|February 9, 2024
Summary
A new method uses visible-near-infrared spectroscopy and image encoding to accurately classify potato varieties. This technique significantly improves accuracy, offering a feasible solution for agricultural processing and consumer needs.
Area of Science:
- Agricultural Science
- Spectroscopy
- Machine Learning
Background:
- Potato variety identification is crucial for processing and consumer satisfaction due to diverse suitability for different applications.
- Existing methods for potato classification may lack efficiency or accuracy for large-scale agricultural needs.
Purpose of the Study:
- To develop and validate a novel method for classifying five distinct potato varieties.
- To enhance potato classification accuracy by integrating visible-near-infrared spectroscopy with advanced image encoding and deep learning techniques.
Main Methods:
- Visible-near-infrared (Vis-NIR) spectra were acquired using transmission and reflection measurements.
- One-dimensional spectral data were transformed into two-dimensional images using Gramian Angular Summation Field (GASF), Gramian Angular Difference Field (GADF), Markov Transition Field (MTF), and Recurrence Plot (RP).
- A ConvNeXt V2 model, enhanced with a coordinated attention mechanism module (ConvNeXt V2-CAP), was utilized for image-based classification.
Main Results:
- Image encoding of spectral data significantly outperformed direct one-dimensional classification models.
- The highest classification accuracy of 99.54% was achieved using GADF image encoding of transmission spectra with the ConvNeXt V2-CAP model.
- The coordinated attention mechanism module improved model performance, particularly on smaller datasets, increasing accuracy by 2.33% with a reduced training set.
Conclusions:
- Visible-near-infrared spectroscopy combined with image encoding technology provides a feasible and highly accurate approach for potato variety classification.
- The proposed ConvNeXt V2-CAP model demonstrates superior performance, offering a robust solution for agricultural applications requiring precise potato identification.
