A Prediction Framework of Apple Orchard Yield with Multispectral Remote Sensing and Ground Features

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Traditional apple yield estimation relies on manual methods, lacking efficiency and multi-source data integration.
• Accurate yield prediction is crucial for agricultural planning and market stability.

Purpose of the Study:

• To develop a novel framework for apple orchard yield prediction integrating multispectral remote sensing and ground features.
• To enable yield prediction at both district/county and small-scale orchard levels.
• To improve the accuracy of apple yield estimation by combining remote sensing with ground-truth data.

Main Methods:

• Extraction of apple orchard planting areas using field-collected samples.
• Large-scale yield prediction using multispectral and environmental features processed via Google Earth Engine and the APYieldNet model.
• Small-scale orchard yield correction using fruit detection (YOLO-A model) and proportional correction methods.

Main Results:

• The proposed APYieldNet model achieved superior performance in large-scale yield prediction (MAE = 152.68 kg/mu, RMSE = 203.92 kg/mu).
• The YOLO-A model demonstrated high accuracy in detecting apple fruits and flowers in complex orchard environments.
• Proportional correction significantly improved the accuracy of small-scale orchard yield predictions, bringing them closer to real yields.

Conclusions:

• The integrated framework effectively predicts apple yields at multiple scales, outperforming existing methods.
• Combining remote sensing data with ground-based fruit detection enhances prediction accuracy.
• This approach offers a more efficient and accurate alternative to traditional manual yield estimation techniques.