TinyML in agriculture: findings and reporting gaps

The review says models typically run inference on-device in the field while training remains in the cloud. Quantization appears in over half of deployments, whereas structured pruning and neural architecture search are rare.

The authors say sparse reporting of flash, RAM, operation counts, and energy per inference hinders reproducibility and fair comparison. They propose an architecture spanning sensing, inference, edge processing, and cloud training.

Why it matters

The review shows why on-device inference may suit small farms with intermittent connectivity, and identifies reporting gaps that make deployments hard to evaluate.