Related Experiment Video
Updated: Aug 30, 2025
06:41
Imaging and Analysis for Quantifying Maize (Zea mays) Abiotic Stress Phenotypes
Published on: March 28, 2025
1000
Nondestructive Classification of Maize Moldy Seeds by Hyperspectral Imaging and Optimal Machine Learning Algorithms
Yating Hu1, Zhi Wang1,2, Xiaofeng Li2,3
1College of Information Technology, Jilin Agricultural University, Changchun 130118, China.
Sensors (Basel, Switzerland)
|August 26, 2022
Summary
This study developed a non-destructive method to detect mildewed maize seeds using hyperspectral imaging (HSI) and a modified random forest (RF) algorithm. The JYSSA-RF approach achieved high accuracy, enabling efficient classification of moldy seeds.
Area of Science:
- Agricultural Science
- Spectroscopy
- Machine Learning
Background:
- Maize seed mildew impacts germination and crop quality.
- Efficient, non-destructive classification of mildewed seeds is crucial.
- Hyperspectral imaging (HSI) offers potential for seed analysis.
Purpose of the Study:
- To establish hyperspectral datasets of maize seeds with varying mildew levels.
- To classify mildewed maize seeds using spectral characteristics and machine learning.
- To optimize a random forest (RF) model with a novel search algorithm.
Main Methods:
- Hyperspectral imaging (HSI) was used to capture spectral data from maize seeds.
- Image processing involved Otus and morphological operations for feature extraction.
- A modified reverse sparrow search algorithm (JYSSA) optimized the RF model.
Main Results:
- The JYSSA-RF algorithm achieved 96% classification accuracy.
- Precision reached 100%, and recall was 93% on the validation set.
- The JYSSA strategy improved the sparrow search algorithm's ability to escape local optima.
Conclusions:
- Hyperspectral imaging combined with JYSSA-optimized RF provides an effective non-destructive method for mildewed seed detection.
- This approach offers novel insights for future seed quality assessment and moldy seed selection.
- The JYSSA algorithm enhances the optimization of machine learning models for agricultural applications.
