Simultaneous corn and soybean yield prediction from remote sensing data using deep transfer learning.

Area of Science:

• Agricultural Science
• Computer Science
• Remote Sensing

Background:

• Remote sensing data enables continuous crop monitoring for large-scale yield estimation.
• Current models predict yield for single crops, lacking simultaneous multi-crop estimation capabilities.
• Accurate yield predictions are crucial for stakeholders to make informed decisions and maximize crop potential.

Purpose of the Study:

• To develop a novel deep learning model for simultaneous estimation of multiple crop yields.
• To incorporate crop interaction effects into yield prediction models.
• To improve the accuracy and efficiency of large-scale crop yield forecasting.

Main Methods:

• A convolutional neural network model, YieldNet, was developed using a novel deep learning framework.
• Transfer learning was employed between corn and soybean yield predictions by sharing backbone feature extractor weights.
• A new loss function was proposed to handle the multi-target response variable in simultaneous yield prediction.

Main Results:

• YieldNet accurately predicts corn and soybean yields 1-4 months before harvest.
• The model achieved Mean Absolute Errors (MAE) of 8.74% for corn and 8.70% for soybean relative to average yields.
• The proposed method demonstrates competitive performance against state-of-the-art approaches in multi-crop yield estimation.

Conclusions:

• YieldNet offers a significant advancement in simultaneous multi-crop yield estimation using remote sensing data.
• The model's ability to consider crop interactions enhances prediction accuracy.
• This approach provides a valuable tool for agricultural stakeholders requiring timely and precise yield forecasts.