SCGNet: efficient sparsely connected group convolution network for wheat grains classification.

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Accurate wheat grain varietal classification is vital for crop yield and disease management.
• Existing methods often face challenges with efficiency and model size.

Purpose of the Study:

• To introduce SCGNet, a novel model for rapid and efficient wheat grain classification.
• To address limitations of traditional and existing machine learning approaches in wheat grain identification.

Main Methods:

• SCGNet utilizes group convolutions for enhanced feature exchange and multiplexing.
• Sparsity in channel connections reduces computational complexity.
• A 3-D convolution-based classification output layer replaces traditional pooling and fully connected layers.

Main Results:

• SCGNet achieved high performance metrics: 99.56% accuracy, 99.59% precision, 99.55% recall, and 99.57% F1-score.
• The model demonstrated superior performance on a curated wheat grain dataset.

Conclusions:

• SCGNet offers a highly efficient solution for wheat grain classification.
• The model's low FLOPs and parameter count make it suitable for practical applications.