Seminar: Graduate Seminar
Gas Sensors Based Smart Agriculture Using TinyML and Reinforcement Learning
Date:
August,17,2025
Start Time:
11:00 - 12:00
Location:
506, Zisapel Building
Add to:
Lecturer:
Adir Krayden
Research Areas:
| Agriculture is undergoing a digital transformation toward Agriculture 4.0, where real-time sensing and intelligent automation are essential for efficiency and sustainability. Smart-farming systems now leverage advanced sensors and AI to monitor crops and produce more closely than ever before. Our research introduces an integrated approach that couples a gas sensor with machine-learning techniques to enable more responsive agricultural decision-making. At its core is GMOS – a miniaturized CMOS-MEMS catalytic gas sensor developed at the Technion, capable of detecting trace volatile compounds such as ethylene, a key ripening hormone. These gases are critical indicators of fruit ripeness, freshness, and storage conditions. By embedding Tiny Machine Learning (TinyML) models directly on the sensor, the system performs on-device gas classification, effectively acting as a โdigital noseโ that distinguishes meaningful emission patterns in real time. Ongoing work is exploring how the same sensing platform could be mounted on a mobile robotic unit. Early benchtop tests and simulation studies suggest that, once matured, such a robot could roam vineyards or storage rooms and make gas-driven decisions on the fly. Reinforcement-learning (RL) control strategies under investigation aim to teach the platform to follow chemical cues laying the groundwork for future autonomous olfactory navigation in produce-handling environments. The envisioned result is a cohesive smart-agriculture platform that can monitor produce condition continuously and adapt its behaviour in real time. Converting gas-sensor data into actionable cues would let growers fine-tune harvest timing and adjust handling or storage practices as soon as quality begins to decline without relying on destructive sampling. Integrating the GMOS sensor with on-board TinyML inference and RL-guided mobility is therefore best viewed as a meaningful step toward more accessible, data-driven tools for precision agriculture.
M.Sc student Under the supervision of Prof. Emeritus Yael Nemirovsky.
|