Developing a TinyML-Oriented Deep Learning Model for an Intelligent Greenhouse Microclimate Control from Multivariate Sensed Data

Equipping the agricultural industry with state-of-the-art technologies such as big data analytics, cloud computing, and the Internet of Things (IoT) is one of the most promising solutions to more efficient and sustainable yield production. In this context, we develop a tinyML-oriented model for a deep learning-based greenhouse microclimate management to be integrated into a on-field microcontroller. Multivariate climate data were collected from sensors installed inside a designed experimental strawberry greenhouse. The obtained values' combinations were labeled according to an eight-action control strategy, then used to prepare a balanced deep learning-ready dataset. The latter was used to train, cross-validate, and test 90 multi-layer perceptrons (MLPs) with varied hyperparameters to select the most performant and optimized model instance for the addressed task. The final selected model incorporates one hidden layer with 8 neurons and has just 104 parameters; it scored a mean accuracy of 95% during the cross-validation phase and 94% on our supplementary test set. The model enables active, robust, and autonomous greenhouse management with the less required computations. It can be efficiently deployed in microcontrollers within real-world operating conditions.