Automated fruit maturity grading using deep learning with feature fusion

Area of Science:

• Agricultural Science
• Computer Science
• Food Science

Background:

• Assessing fruit ripeness is challenging due to complex physiological and environmental factors.
• Visual inspection is unreliable for certain mango varieties like Nam Dok Mai Si Tong (NDMST) due to uniform skin color during ripening.
• Current methods for fruit maturity grading often lack accuracy and practicality for real-world applications.

Purpose of the Study:

• To develop an automated fruit maturity grading system for NDMST mangoes.
• To integrate external visual features with internal biochemical attributes for enhanced ripeness assessment.
• To create a practical, non-destructive, and cost-effective solution for automated fruit grading.

Main Methods:

• Proposed AFMG-DLFF (Automated Fruit Maturity Grading using Deep Learning with Feature Fusion), a multimodal deep learning framework.
• Extracted visual features using DenseNet201, Inception-ResNetV2, and EfficientNetV2.
• Encoded biochemical traits (total soluble solids, titratable acidity, BrimA) via a dedicated neural network.
• Fused and optimized feature spaces using Glowworm Swarm Optimization (GSO) for hyperparameter tuning.
• Trained the model with an 80:20 train-test split and early-stopping-based validation.

Main Results:

• Achieved a classification accuracy of 97.86% for NDMST mango maturity stages.
• Demonstrated strong performance compared to deep-learning baselines.
• Remained competitive with existing literature-reported fruit ripeness classification methods.
• Utilized accessible RGB imaging and standard biochemical measurements.

Conclusions:

• The AFMG-DLFF framework provides a highly accurate method for automated fruit maturity grading.
• The multimodal approach effectively combines visual and biochemical data for improved ripeness assessment.
• This technology offers a practical, non-destructive, and cost-effective solution for the fruit supply chain.