New imaging technology paves the way for simplified, low-cost agricultural quality assessment

Hyperspectral imaging is a technique that helps analyze the chemical composition of foods and agricultural products. However, this is an expensive and complicated procedure, which limits its practical application.

A team of researchers at the University of Illinois at Urbana-Champaign has developed a method to reconstruct hyperspectral images from standard RGB images using deep machine learning. This technology greatly simplifies the analytical process and has the potential to revolutionize product evaluation in the agricultural industry.

“Hyperspectral imaging uses expensive equipment. Being able to use RGB images taken with regular cameras or smartphones allows us to predict product quality using low-cost handheld devices,” said first author. said Md Toukir Ahmed.

Ahmed is a doctoral student in the College of Agricultural and Biological Engineering (ABE), part of the Illinois College of Agricultural, Consumer, and Environmental Sciences and Granger Institute of Technology.

The researchers tested their method by analyzing the chemical composition of sweet potatoes.

They focused on soluble solids in one study (published in the Journal of Food Engineering) and dry matter in the second study (published in Results in Engineering). This is an important characteristic that influences the taste, nutritional value, marketability, and processing suitability of sweeteners. Potato.

A deep learning model was used to convert the information in an RGB image into a hyperspectral image.

“RGB images can only detect visible attributes such as color, shape, size, and external defects; chemical parameters cannot be detected. , and blue.

“But in hyperspectral images, there are many channels and wavelengths from 700 to 1,000 nm. Using deep learning techniques, we can map and reconstruct that range, making it possible to detect chemical attributes from RGB images. ” said Mohamed Kamruzzaman, assistant professor at ABE. and corresponding author of both papers.

Hyperspectral imaging captures detailed spectral signatures at spatial locations across hundreds of narrow bands and combines them to form a hypercube. Applying state-of-the-art deep learning-based algorithms, Kamruzzaman and Ahmed were able to create a model that reconstructs hypercubes from RGB images, providing relevant information for product analysis.

They used reconstructed hyperspectral images of sweet potatoes to calibrate a spectral model and achieved prediction accuracy of over 70% for soluble solids content and over 88% for dry matter content, extending previous studies. showed a significant improvement compared to

In the third paper, published in the journal Smart Agriculture Technology, the researchers applied deep learning techniques to reconstruct hyperspectral images to predict mortality in chicken embryos. This can be applied to the egg and hatchery industry. They considered various techniques and recommended the most accurate approach.

“Our results show great potential to revolutionize the quality assessment of agricultural products. Reconstructing detailed chemical information from simple RGB images opens new lines of analysis that are both affordable and accessible. Possibilities are open.

“While challenges remain in scaling this technology to industrial applications, this is a truly exciting initiative as it has the potential to transform quality control across the agricultural sector,” concluded Kamruzzaman. Ta.