TinyML for Real-Time Crop Disease Detection in Precision Agriculture: A Proof-of-Concept Study on Maize Leaf Disease Detection

Traditional machine learning approaches for crop disease detection often require substantial computational resources, energy consumption, and net-work connectivity. Tiny Machine Learning (TinyML) addresses these limita-tions by enabling machine learning inference directly on resource-constrained edge devices, like microcontrollers. It is therefore well-suited for remote field applications with limited or nonexistent internet connectivity and cloud infrastructure. This study proposes a TinyML-based framework for real-time crop disease detection, demonstrated through a proof-of-concept study on binary maize leaf disease detection. Three lightweight deep learning models, namely a custom CNN, a custom MobileViT model, and a transfer learning-based MobileNetV1 model, were developed and evaluated on two publicly available maize leaf image datasets. To ensure a robust performance assessment, five-fold cross-validation was used, and model performance was evaluated using accuracy, precision, recall, F1-score, false positives, and false negatives. The trained models were subsequently opti-mised and converted to TensorFlow Lite format for deployment on Arduino Portenta H7. The comparative analysis showed that while MobileNetV1 demonstrated superior deployment efficiency with a latency of 71 ms and RAM of 125 KB, custom CNN and transformer-based MobileViT models achieved mean classification accuracy of 91.90% and 96.07%, comparable to, and in some cases exceeding, that of the transfer learning-based MobileNetV1 model. The results also demonstrate the trade-off between disease detection performance and deployment feasibility, highlighting the need to balance predictive accuracy with computational efficiency in TinyML applications. Using binary maize leaf disease detection as a proof-of-concept, this study demonstrates the feasibility of deploying lightweight deep learning models on resource-constrained edge devices for real-time crop disease detection in precision agriculture.