Detection of sweet corn seed viability based on hyperspectral imaging combined with firefly algorithm optimized deep...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Sweet corn seed vitality is crucial for crop yield and quality.
• Accurate identification of seed viability is essential for breeding and agriculture.
• Traditional methods for assessing seed vitality can be time-consuming and labor-intensive.

Purpose of the Study:

• To develop and evaluate deep learning models for identifying sweet corn seed vitality grades.
• To compare the performance of various deep learning algorithms against traditional machine learning methods.
• To introduce a novel Firefly Algorithm optimized CNN-LSTM model for enhanced seed vitality classification.

Main Methods:

• Hyperspectral imaging was used to collect spectral data from 496 sweet corn seeds across four viability grades.
• Support Vector Machine (SVM) and Extreme Learning Machine (ELM) were employed as baseline machine learning models.
• Deep learning models including 1DCNN, 1DLSTM, CNN-LSTM, and the proposed FA-CNN-LSTM were developed and tested.

Main Results:

• Deep learning models significantly outperformed traditional machine learning, achieving over 94.26% accuracy.
• The proposed FA-CNN-LSTM model demonstrated superior performance, reaching a classification accuracy of 97.23%.
• FA-CNN-LSTM showed a 1.49% improvement over the standard CNN-LSTM model and a 2.97% improvement over the lowest-performing CNN.

Conclusions:

• Integrating deep learning with hyperspectral imaging offers a powerful and accurate method for discriminating sweet corn seed vitality.
• The FA-CNN-LSTM model presents a promising advancement for automated seed quality assessment in agriculture.
• This approach has significant potential for application in agricultural research and cultivar breeding programs.