Design and evaluation of an intelligent sorting system for bell pepper using deep convolutional neural networks

Area of Science:

• Agricultural Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Bell pepper quality homogeneity is crucial for consumers and the food industry.
• Current sorting methods may not meet the precision required by export standards for maturity and size.

Purpose of the Study:

• To develop an automated in-line sorting system for bell peppers.
• To grade bell peppers into five classes based on export criteria using deep learning.

Main Methods:

• A deep convolutional neural network (DCNN) model, specifically ResNet50, was adapted with a custom classifier.
• The model was trained and validated using a five-fold cross-validation approach.
• The DCNN model was integrated into a machine vision-based sorting machine for real-time application.

Main Results:

• The developed DCNN model achieved high performance metrics in the in-line phase: 98.7% accuracy, 97% precision, 96.9% sensitivity, 99% specificity, and 96.9% F1-score.
• The system processed approximately 3000 bell pepper samples per hour with a single sorting line.
• Classification time per sample was as low as 4 ms, enabling real-time processing.

Conclusions:

• The proposed DCNN-integrated sorting system effectively grades bell peppers according to export standards.
• The system demonstrates high accuracy and efficiency, suitable for industrial-scale agricultural applications.
• This intelligent system offers a viable solution for automated quality control in the bell pepper industry.