🌿 Context-Aware Deep Learning Model for Yield Prediction Using Time-Series UAS Multispectral Data

📖 About the Research

This reserach integrates time-series uncrewed aerial system (UAS) multispectral imaging with advanced deep learning techniques to enhance field-scale crop yield prediction. Using a MicaSense RedEdge MX+ sensor, multispectral data was collected across different nitrogen (N) rates throughout the potato growing season. The dataset was preprocessed and analyzed using a combination of Partial Least Squares Regression (PLSR) and a novel deep learning architecture (CAR Conv1D-BiGRU-BiLSTM-Net) to model temporal crop dynamics effectively.

🏗 Research Pipeline

The pipeline for UAS-based crop yield prediction follows a structured workflow:

🚀 Features

• 🌍 Multispectral UAS data analysis with Pix4Dmapper, QGIS, and linear unmixing techniqe.
• 📈 Feature extraction using PLSR for robust prediction
• 📊 Time-series crop yield prediction using deep learning
• 🔄 Growth stage-based yield modeling (T1-T5)
• 🖥️ Python-based implementation with geopandas, rasterio, sklearn, TensorFlow & SciPy

📂 Dataset & Preprocessing

• Data collected using UAS multispectral imaging
• Preprocessing: Pix4Dmapper, QGIS, Linear unmixing, and soil masking techniques
• Feature engineering: Vegetation indices (SR, CHLGR, MARI, VF), N-rate

🧠 Deep Learning Model

• PLSR for feature extraction relevant to crop yield
• CAR Conv1D-BiGRU-BiLSTM-Net for time-series prediction
• Outperformed other models with R² = 0.775 and RMSE = 16.4%

🛠 Installation

1. Clone the repository:

   git clone https://github.com/SAY70/CAR-CNN-BiRNN.git
   cd CAR-CNN-BiRNN

📖 Citation

@article{yadav2025context,
title={Context-Aware Deep Learning Model for Yield Prediction in Potato Using Time-Series UAS Multispectral Data},
author={Yadav, Suraj A and Zhang, Xin and Wijewardane, Nuwan K and Feldman, Max and Qin, Ruijun and Huang, Yanbo and Samiappan, Sathishkumar and Young, Wyatt and Tapia, Francisco G},
journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
year={2025},
publisher={IEEE}
}

🤝 Contributing

We welcome contributions! Feel free to submit pull requests or open issues.

---

🌟 Star this repository if you find it useful!