Related Experiment Videos
Low-cost embedded system for spectral power distribution reconstruction for controlled environmental agriculture
Juan Morales-Guerra1, Juan Soto-Perdomo2, Juan Botero-Valencia1
1Faculty of Engineering, Instituto Tecnológico Metropolitano, Grupo SCR, Medellín 050034, Colombia.
Hardwarex
|June 22, 2026
Summary
An open-source device uses machine learning to accurately measure LED spectral power distribution (SPD) for controlled agriculture. This cost-effective solution improves spectral monitoring accuracy in horticultural lighting systems.
Area of Science:
- Agricultural Engineering
- Spectroscopy
- Machine Learning
Background:
- Accurate spectral power distribution (SPD) measurement is crucial for optimizing LED lighting in controlled environmental agriculture (CEA).
- Traditional spectrometers can be expensive and cumbersome for widespread deployment in agricultural settings.
- Developing cost-effective, accurate spectral monitoring solutions is essential for advancing horticultural practices.
Purpose of the Study:
- To develop and validate an open-source, low-cost device for acquiring, correcting, and reconstructing the SPD of LED sources in CEA.
- To implement a machine learning-based pipeline for inferring dense SPD from sparse multispectral sensor data.
- To integrate environmental monitoring for correlating spectral anomalies with environmental conditions.
Main Methods:
- Utilized a low-cost AS7265x multispectral sensor (18 channels, 410-940 nm) for sparse spectral data acquisition.
- Developed a two-stage machine learning pipeline: a multilayer perceptron (MLP) for in-device correction and a 1D-CNN for cloud-based spectral reconstruction.
- Integrated a BME688 sensor for environmental monitoring (temperature, humidity, gas concentration).
Main Results:
- The MLP correction stage significantly reduced the RMSE from 0.183 to 0.035, enhancing data reliability.
- The 1D-CNN spectral reconstruction model achieved a high precision with an RMSE of 0.0135 for horticultural LED spectra.
- The system demonstrated effective wireless data transmission and remote visualization capabilities.
Conclusions:
- The developed open-source device offers a scalable and cost-effective solution for spectral monitoring in CEA.
- Machine learning significantly improves the accuracy and reliability of spectral measurements from low-cost sensors.
- This technology has the potential to optimize LED lighting strategies and improve crop yields in controlled agricultural environments.
