Enhanced Infield Agriculture with Interpretable Machine Learning Approaches for Crop Classification

Explainability is critical for AI in agriculture: The paper identifies "the need for model explainability" as the most significant limitation in existing crop classification research, highlighting that farmers and agricultural stakeholders need to understand why an AI system makes its predictions, not just accept its outputs.

Simpler models can outperform complex ones: Xception, a transfer learning model, achieved 98% accuracy while remaining practical for real-world use with an 80 MB file size and 0.0633-second prediction time—demonstrating that the most sophisticated AI approaches aren't always necessary for effective crop classification.

Multiple AI approaches should be evaluated together: The research compared traditional machine learning, custom deep learning, established architectures, and cutting-edge foundation models, finding that "all models performed well"—suggesting that responsible AI deployment requires testing diverse approaches rather than defaulting to the newest technology.

Model transparency enables agricultural adoption: By applying Explainable AI techniques to interpret model decisions, this research addresses a practical barrier to farmers trusting and adopting AI tools in their operations, making the technology more accessible to non-technical users in the field.