Machine Learning Modeling of Mid-Infrared Spectroscopy Data for Rapid Sample Analysis

Executive Statement:

The technology leverages mid-infrared spectroscopy data and machine learning modeling to provide multi-property analysis that is fast and cost-effective, compared to traditional wet chemistry testing methods.

Description:

The MSU research team developed machine learning models using mid-infrared (MIR) spectroscopy data for fast and cost-effective analysis of diverse properties. The researchers used soil data (Kellogg Soil Survey Laboratory database, Natural Resources Conservation Service, USDA; Soil, Plant, and Nutrient Lab database, Michigan State University) to develop models to predict key soil properties based off the MIR spectroscopy data of a sample, instead of using multiple wet chemistry tests to obtain the same soil properties. For soils, the technology predicts key soil properties such as total carbon, organic matter, pH, and nutrient content. While the technology was developed using soil data, the technology is applicable to other sample types (e.g., food, edible oil, feed, growth media). The technology offers a rapid and accurate alternative to traditional testing methods for multiple sample types and properties.

Key Advantages:

• Faster and cheaper analysis compared to traditional wet chemistry methods
• Capable of predicting multiple properties simultaneously
• Flexible – can be used for different sample types and properties

Applications:

• Soil quality and agriculture fertility management
• Environmental monitoring and management  
• Soil carbon monitoring and management for carbon credit purposes
• Food and feed testing and manufacture monitoring

Patent Status: US Provisional Patent Application filed

Licensing Rights: Full licensing rights available

Selected Publications:

Developing Spectral Libraries Using Mid Infrared Spectroscopy to Determine Key Soil Properties and Soil Health, Faisal Sherif, 2023.30633911 (Michigan State University Master’s Thesis). ProQuest

Inventors: Faisal Sherif, Jonathan Dahl, and Jessica Miesel