Adaptive IoT-enabled automation with multimodal sensing and real-time image analysis for CEA
Authors
Arshad Kadevalappil Ajilan (presenting author) [1]
Affiliations
- Grønt fra Nord AS, Fauske, Norway
Email (presenting author): arshad@grontfranord.no
Keywords
controlled environment agriculture (CEA); IoT automation; multimodal sensing; image-based phenotyping; machine learning; predictive control; smart farming
Abstract
Sustainable controlled environment agriculture (CEA) requires precise environmental regulation, yet manual control of lighting, nutrients and climate remains labor-intensive and difficult to optimize at commercial scale. We are developing an adaptive control framework using ESP32-based sensor nodes and camera modules to monitor physical, chemical and visual plant parameters in real time and support closed-loop environmental regulation. Temperature, humidity, VOCs, CO₂, EC, pH and RGB spectral measurements are collected alongside plant images from an ESP32-CAM. An ESP32 microcontroller functions as the central controller, transmitting data to a cloud computer for multivariate analysis and using image-derived phenotypic indicators to inform actuation of fans, lights, CO₂ injectors and nutrient dosing. To demonstrate feasibility, we developed an MVP prototype that integrates temperature, humidity and image data to regulate LED growth lighting (including brightness and channel combinations) and fan speed. Current image analysis quantifies leaf color metrics and applies a general vision model to derive a proxy health indicator, which is integrated with sensor data for feedback-driven light adjustment. All measurements are transmitted via MQTT and stored in a MySQL database for evaluation and system refinement. Future work will integrate feature extraction, statistical inference and predictive modeling to identify patterns within the commercial growth system at Grønt fra Nord and forecast yield constraints such as etiolation and disease-associated stress signals. Machine and deep learning models trained on accumulated data will capture nonlinear interactions between environmental variables and plant responses, enabling predictive control strategies that reduce the risk of suboptimal growth conditions. Overall, this work outlines a practical pathway toward data-driven, closed-loop environmental control in CEA, with the goal of improving operational sustainability.
Funding
This work was supported by internal research and development funds from Grønt fra Nord AS.