Novel transfer learning approach for detecting mango fruit type and quality assessment

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Mango varieties differ significantly in flavor, texture, and appearance, necessitating precise classification.
• Effective supply chain management and meeting consumer demands rely on accurate mango variety identification and quality assessment.

Purpose of the Study:

• To develop an innovative methodology for classifying mango varieties and evaluating their quality using transfer learning and machine learning.
• To introduce the IncepForestNet approach for feature engineering in mango classification and quality assessment.

Main Methods:

• Leveraged pre-trained Inception V3 models via transfer learning, fine-tuned with a mango image dataset.
• Employed the novel IncepForestNet for spatial feature extraction, followed by Random Forest for probabilistic feature identification.
• Integrated and compared various machine learning algorithms for classification and quality assessment.

Main Results:

• The proposed approach accurately classifies mango varieties and assesses key quality attributes.
• The Random Forest model demonstrated outstanding performance with 99% accuracy and robust k-fold validation.
• Comparative analysis identified optimal machine learning techniques for mango classification and quality assessment.

Conclusions:

• The study presents a robust methodology combining transfer learning and machine learning for mango classification and quality evaluation.
• The developed approach has significant potential to optimize agricultural supply chains and enhance consumer satisfaction.
• The Random Forest model proves highly effective for both mango variety classification and quality assessment tasks.