Low-cost NIR spectroscopy sensor for on-site soil nitrogen and carbon detection

Overview

This project develops a low-cost near-infrared (NIR) spectroscopy sensor for on-site detection of soil nitrogen (N) and carbon (C) content. By combining digital light processing (DLP) technology with calibration models built using the back-propagation neural network (BPNN) algorithm, the sensor aims to deliver rapid, accurate, and affordable soil nutrient analysis directly in the field. The solution targets precision agriculture, environmental monitoring, and soil health management, enabling farmers, agronomists, and land managers to make data-driven decisions without relying on laboratory-based testing. Compared to existing commercial NIR spectroscopic sensors, the proposed device is designed to be approximately 50% cheaper or more, broadening access to advanced soil analysis for a wider range of users.

Technical specifications

• DLP-based NIR spectroscopy sensor: Uses digital light processing technology to reduce hardware costs while maintaining spectral measurement capability for soil analysis.
• BPNN calibration models: Back-propagation neural network algorithms are used to establish spectral calibration models that correlate NIR spectral data with soil nitrogen and carbon content, achieving high prediction accuracy under both dry and wet soil conditions.
• Multi-parameter detection: Designed to simultaneously detect soil nitrogen and carbon content on-site, extending prior laboratory-validated single-parameter models to a multi-parameter field application.
• Validated prediction performance: Laboratory validation using 459 sets of soil data demonstrated excellent prediction accuracy, with R-squared values of 0.99 for dry soil samples and 0.93 for wet soil samples for soil nitrogen content.
• Wet and dry soil compatibility: Calibration models perform reliably across varying soil moisture conditions, supporting practical field deployment where soil moisture levels fluctuate.

Technology readiness level

The technology is currently at an advanced laboratory validation stage. Calibration models using the BPNN algorithm have been validated with 459 soil sample datasets, demonstrating high prediction accuracy for soil nitrogen content under both dry and wet conditions, and showing potential for extension to soil carbon detection. The next phase involves developing the DLP-based low-cost NIR sensor hardware, establishing new calibration models for soil carbon and nitrogen, and integrating the models into the sensor for on-site multi-parameter detection. The planned validation timeline is approximately one year. The project is seeking partnerships to support low-cost sensor development, spectral calibration model refinement, soil testing equipment, and funding for key personnel.