WITCH

WITCH-CRAFT en agricultura: aplicación de la AIoT para un cultivo más competitivo y saludable (WITCH)

Arthur C. Clarke wrote that “any sufficiently advanced technology is indistinguishable from magic.” WITCH takes its name from this idea. Its goal is to bring a more advanced, efficient and low-power artificial intelligence to the Andalusian agricultural sector, even if it may look like “witchcraft” to its future users.

Agriculture faces climate change, desertification and the need to cut the use of water, fertilisers, pesticides and energy. A small error in decision-making can also compromise a whole season. WITCH proposes combining the Internet of Things (IoT) and Artificial Intelligence (AI), known as AIoT, under the Cloud-to-Things Continuum paradigm. The aim is a smart, low-cost system for monitoring and acting on crops. Data from multiple sensors are processed directly in the field (Edge Computing) using ultra-low-power learning architectures such as Spiking Neural Networks (SNN) and Spiking Neural P Systems (SNPS). This greatly reduces the amount of data to transmit and extends battery life. More complex models run at higher levels (Fog and Cloud), where data from several plantations and seasons are combined.

The system is based on four subsystems:

  1. AIoT node. An autonomous device installed next to the plants and powered by batteries and solar energy. It integrates the WITCH sensors: Weather, Insects (image-based insect detection), Temperature, Colour (colour and light) and Humidity (soil moisture). Its core is a reconfigurable system-on-chip (MPSoC with ARM and FPGA) that runs the AI models locally. It sends only the plant’s status, needs and alarms over 802.15.4 or LoRa. In parallel, a custom RISC-V-based ASIC will be designed in TSMC 65 nm technology, in line with the Spanish PERTE Chip initiative.
  2. Router. The gateway between the node network and the cloud over mobile networks (4G/5G). It can store data, process it at Fog level or forward it to the Cloud. It will also help find the best placement for nodes and router by generating a coverage map as the user walks the plot.
  3. Mobile app. Lets farmers and agronomists check crop status and metrics through a conversational interface based on language models. It includes visualisations of cultivation processes obtained through process mining.
  4. Capacity- and cost-aware governance. This infrastructure dynamically decides where each service runs along the edge-fog-cloud continuum, based on the energy consumption and cost of each component.

Objectives

  • O1: Prototype of the AIoT node and router. (Modules 1 & 4)
  • O2: Low-power AI models for Edge AI (SNN / SNPS). (Module 2)
  • O3: Edge-fog-cloud software architecture with a mobile app. (Module 3)
  • O4: Governance infrastructure based on capacity and costs. (Module 5)
  • O5: Validation through field-test campaigns in real plantations. (Module 6)

First results

A preliminary node prototype has already been built. It includes soil moisture and temperature, UV radiation, IMU, GPS, ambient humidity and VGA/IR camera sensors, plus LoRa communication. Its enclosure has also been designed. The first tests were carried out in the IRNAS-CSIC greenhouses. Work is also under way on monitoring water stress in a super-intensive olive grove in Utrera (Seville), using dendrometers, sap-flow sensors and leaf turgor-pressure sensors.


Principal Investigators: Alejandro Linares Barranco (PI1) and José María García Rodríguez (PI2)
Project type: University R&D&I project (EIDIA), Line 2: applied research. Junta de Andalucía.
Reference: DGP_PIDI_2024_00918
Start date: May 2026
End date: April 2029
Group: TEP108 – Robotics and Computer Technology. SCORE Unit of Excellence, University of Seville.

Researchers:
Claudio Amaya Rodríguez
José Javier Berrocal Olmeda
Elena Cerezuela Escudero
Margarita Cruz Risco
Irene Bedilia Estrada Torres
José Manuel García Alonso
Francisco de Asís Gómez Rodríguez
Ángel Francisco Jiménez Fernández
Octavio Martín Díaz
Carlos Guillermo Müller Cejás
Juan Manuel Murillo Rodríguez
David Orellana Martín
Manuel Resinas Arias de Reyna
José Antonio Ríos Navarro

Work team:
José Antonio Andreu Guzmán, Matthew Bwye Lera, Daniel Casanueva Morato, Dagnier Antonio Curra Sosa, Santiago Díaz Romero, María Elena Molino Peña, Juan Manuel Montes Sánchez, Tomás Muñoz González, Antonio Manuel Pérez Peña, David Ragel Díaz-Jara, Manuel Rivas Pérez, Pablo Sánchez Cuevas, Diógenes Enrique Trejos Hernández.

Collaborating entities: BASF (Estación Experimental Agrícola de Utrera), IRNAS-CSIC (Finca experimental «La Hampa», Coria del Río), COBER, Midway Tech y LifeWatch ERIC.

This work was funded by the Junta de Andalucía (Consejería de Universidad, Investigación e Innovación) and co-funded by the EU through the FEDER Andalucía 2021–2027 Programme under project DGP_PIDI_2024_00918.