Predicting and Visualizing Citrus Color Transformation Using a Deep Mask-Guided Generative Network.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Citrus rind color is a key indicator of fruit maturity and development.
• Accurate monitoring and prediction of rind color changes are crucial for optimizing crop management and harvest timing.

Purpose of the Study:

• To develop and validate a high-accuracy workflow for predicting and visualizing citrus rind color transformation.
• To create a versatile model capable of forecasting rind color at various future time points.

Main Methods:

• A novel framework integrating visual saliency with deep learning, comprising segmentation, generative, and loss networks.
• Fusion of image features and temporal information within a single model to predict color changes over time.
• Development of a dataset with 7,535 images from 107 Navel oranges during color transformation.

Main Results:

• The semantic segmentation network achieved a mean intersection over union score of 0.9694.
• The generative network demonstrated high image quality and similarity, with a peak signal-to-noise ratio of 30.01.
• The model's predictions are consistent with human perception and validated by quantitative metrics.

Conclusions:

• The proposed AI framework accurately predicts and visualizes citrus rind color transformation.
• The model's efficiency and accuracy are suitable for real-world applications, including a mobile app.
• The methodology is adaptable for predicting color changes in other fruit crops.