Knowledge-guided adaptive spatial-temporal graph contrastive learning framework: Regional crop diseases prediction...

Area of Science:

• Agricultural Science
• Data Science
• Computer Science

Background:

• Crop disease prediction faces challenges due to complex dynamics and data limitations.
• Existing smart agriculture methods struggle with real-time, accurate regional forecasting.
• Plant Electronic Medical Records (PEMRs) offer a novel big data source for disease analysis.

Purpose of the Study:

• To develop a robust framework for real-time regional crop disease prediction.
• To leverage Plant Electronic Medical Records (PEMRs) for enhanced spatial-temporal forecasting.
• To address limitations in data acquisition, cost, and disease transmission complexity.

Main Methods:

• Proposed a knowledge-guided adaptive spatial-temporal graph contrastive learning framework (KAST-Graph).
• Quantified regional disease occurrence and modeled it as spatial-temporal graph forecasting.
• Integrated adaptive, geographically-informed adjacency matrices and a contrastive learning enhancement scheme.

Main Results:

• KAST-Graph demonstrated superior performance over state-of-the-art baselines.
• Achieved excellent spatial-temporal mining results on PEMRs big data.
• Reported best MAE (5.71), RMSE (9.50), and MAPE (4.56 %) scores.

Conclusions:

• KAST-Graph significantly advances regional crop disease prediction capabilities.
• The framework enhances robustness against noisy and incomplete data.
• This research provides critical tools for early warning and prevention in smart agriculture.