Mobile, low-cost multispectral sensing system for estimating nitrogen levels in maize

Overview

This solution offers a portable, low-cost multispectral sensing system designed to estimate nitrogen (N) levels in maize leaves directly in the field. By pairing a compact multispectral sensor with a mobile smartphone application powered by machine learning, the system delivers rapid, non-destructive nutrient diagnostics without requiring expensive laboratory equipment or hyperspectral imaging infrastructure. The technology addresses a critical need in precision agriculture: timely assessment of crop nitrogen status to guide fertilizer decisions, improve yields, and reduce environmental impact from over-application.

The system is being developed by the International Fertilizer Development Center (IFDC), a global applied research organization with deep expertise in nutrient management, greenhouse trial capabilities, and pathways to farmer adoption through international partnerships.

Technical specifications

Core components:

• A low-cost multispectral sensor capturing 18 spectral channels across the 400 nm to 950 nm range, the spectrum where nitrogen deficiency produces distinctive absorption signatures in maize leaves
• An integrated electronic board with LED lighting control and mobile communication interface for consistent, repeatable leaf scanning
• A 3D-printed enclosure that isolates the leaf sample from ambient light, ensuring measurement reliability in field conditions
• A neural network algorithm trained on greenhouse trial data with varying nitrogen uptake levels to predict leaf N status
• A mobile application built using the TensorFlow Lite API that runs the trained model on-device for real-time inference

How it works:

• Nitrogen levels in maize leaves correlate strongly with chlorophyll concentration, which in turn affects spectral reflectance patterns
• The multispectral sensor captures reflectance data across visible and near-infrared wavelengths where N-deficient and healthy plants show distinguishable signatures
• The neural network model processes this spectral data to classify or estimate nitrogen levels on-site, eliminating the need for laboratory analysis

Technology readiness level

The technology is currently at an early-to-mid development stage. The research hypothesis and validation procedures are established, with greenhouse trials planned to generate training data across different nitrogen regimes. The multispectral hardware configuration and machine learning pipeline have been designed but require full laboratory and field validation. The next steps include assembling the prototype device, conducting greenhouse trials to train the neural network, porting the model to a mobile application, and validating accuracy against known nitrogen levels. Commercial deployment readiness will depend on the outcomes of these validation activities and subsequent field testing across diverse growing conditions.