Computer Vision in Monitoring Fruit Browning: Neural Networks vs. Stochastic Modelling

Area of Science:

• Agricultural Science
• Computer Vision
• Food Quality Control

Background:

• Conventional fruit browning inspection is subjective, time-consuming, and costly.
• Computer vision offers an automated solution for monitoring and sorting in the agrifood sector.
• Enzymatic browning in pears impacts quality and shelf-life, necessitating efficient detection methods.

Purpose of the Study:

• To investigate computer vision techniques for detecting and classifying enzymatic browning in cut pears.
• To compare the performance of convolutional neural networks (CNNs) with stochastic modeling for this task.
• To explore the potential of a hybrid approach for enhanced automated quality control.

Main Methods:

• Application of various RGB cameras and computer vision algorithms.
• Development and training of a CNN model for image analysis.
• Implementation of stochastic modeling using the CIE Lab* color model to derive Browning Index (BI) and Yellowing Index (YI).
• Integration of BI and YI with a Bayesian classifier.

Main Results:

• The developed CNN model achieved 96.6% accuracy and an F1-score >0.96 for pear slice testing.
• Stochastic modeling provided quantitative BI and YI for precise browning monitoring.
• Combining BI and YI with a Bayesian classifier improved classification rates by 4.6% for control and 15% for treated samples.

Conclusions:

• CNNs offer high-throughput, adaptable solutions for automated browning detection.
• Stochastic methods provide precise, quantitative analysis but require human expertise.
• A hybrid approach combining CNNs and stochastic modeling is recommended for robust and practical image analysis in agricultural quality control.