Two-Stage Intelligent DarkNet-SqueezeNet Architecture-Based Framework for Multiclass Rice Grain Variety...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Manual identification of rice varieties is labor-intensive and time-consuming.
• Accurate crop variety identification is vital for agriculture and food security.
• Existing methods may lack the precision required for large-scale rice classification.

Purpose of the Study:

• To develop an automated, accurate system for detecting and classifying multiclass rice grain varieties.
• To leverage deep learning and optimization algorithms for enhanced agricultural image analysis.
• To improve the efficiency and reliability of rice variety identification.

Main Methods:

• A two-stage deep learning framework utilizing Darknet19 and SqueezeNet models.
• Transfer learning for fine-tuning pretrained models on a rice grain dataset.
• Feature extraction, fusion via maximum correlation, and selection using an improved butterfly optimization algorithm (BOA).
• Classification using various machine learning algorithms.

Main Results:

• The proposed framework achieved a maximum classification accuracy of 100% on five rice varieties.
• The average accuracy across experiments was 99.2%, demonstrating high reliability.
• The improved BOA effectively selected optimal features, enhancing classification performance.

Conclusions:

• The developed deep learning framework offers a highly accurate and efficient solution for multiclass rice grain variety detection.
• The integration of advanced feature selection techniques significantly boosts classification performance.
• This research provides a valuable tool for agricultural applications requiring precise crop identification.