Exploring Deep Learning Approaches for Walnut Phenotype Variety Classification.

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Efficient classification of agricultural commodities is vital for quality assessment and supply chain management.
• Walnut classification presents challenges due to varietal differences and the need for accurate quality grading.

Purpose of the Study:

• To analyze deep learning and data science methods for classifying walnut varieties.
• To evaluate the performance of different deep learning architectures and machine learning classifiers for walnut image classification.

Main Methods:

• Collected a dataset of walnut images from Chandler, Fernor, Howard, and Oguzlar varieties.
• Conducted two experiments: 1) using deep learning models (InceptionV3, VGG-19, VGG-16) as direct classifiers, and 2) using deep learning for feature extraction followed by Support Vector Machine (SVM), Logistic Regression (LR), and k-Nearest Neighbors (k-NN) classifiers.

Main Results:

• In the first experiment, InceptionV3 achieved the highest classification accuracy, outperforming VGG-19 and VGG-16.
• The second experiment, utilizing deep learning for feature extraction, showed improved overall success rates.
• The combination of InceptionV3 for feature extraction and Logistic Regression for classification yielded the highest success rate.

Conclusions:

• Deep learning methodologies are highly effective for rapid and accurate agricultural product classification based on visual data.
• The findings suggest potential for enhancing agricultural classification systems through advanced computational techniques.
• The InceptionV3 and Logistic Regression model offers a promising approach for automated walnut quality assessment.