University of Illinois Urbana-Champaign researchers have pioneered a novel approach using deep learning to convert standard RGB images into hyperspectral images. This innovative technique could drastically simplify and reduce the costs of agricultural quality assessments.
In a significant technological advancement, a team of researchers from the University of Illinois Urbana-Champaign has devised a method to reconstruct hyperspectral images from simple RGB images using deep machine learning. This innovative approach promises to simplify and reduce the cost of agricultural quality assessments, potentially revolutionizing the industry.
“Hyperspectral imaging uses expensive equipment. If we can use RGB images captured with a regular camera or smartphone, we can use a low-cost, handheld device to predict product quality,” lead author Md Toukir Ahmed, a doctoral student in the Department of Agricultural and Biological Engineering (ABE), said in a news release.
Hyperspectral imaging, although highly effective for analyzing the chemical composition of food and agricultural products, has historically been a costly and complex procedure. Conventional methods require specialized equipment to capture detailed spectral signatures across hundreds of narrow bands, forming hypercubes that provide intricate chemical information. This complexity has limited its practical application in the industry.
Using deep learning algorithms, the team led by Ahmed and Mohammed Kamruzzaman, an assistant professor in ABE, developed a model to transform RGB images into hyperspectral ones.
The team validated their model by analyzing sweet potatoes, focusing on soluble solid content in one study and dry matter in a second study — key indicators of taste, nutritional value and marketability. Their method achieved over 70% accuracy in predicting soluble solid content and 88% accuracy in dry matter content, outperforming previous studies significantly.
“With RGB images, you can only detect visible attributes like color, shape, size and external defects; you can’t detect any chemical parameters,” Kamruzzaman, a corresponding author on both studies, said in the news release. “With deep learning methods, we can map and reconstruct that range so we now can detect the chemical attributes from RGB images.”
Furthermore, their research extended beyond plant products.
In another study, the deep learning techniques were applied to predict chick embryo mortality, demonstrating potential applications in the egg and hatchery industry. This versatility underscores the wide-ranging implications of their approach.
“Our results show great promise for revolutionizing agricultural product quality assessment,” added Kamruzzaman. “By reconstructing detailed chemical information from simple RGB images, we’re opening new possibilities for affordable, accessible analysis. While challenges remain in scaling this technology for industrial use, the potential to transform quality control across the agricultural sector makes this a truly exciting endeavor.”
The breakthrough signifies a monumental step toward making advanced hyperspectral imaging accessible and affordable, potentially transforming how quality assessments are conducted in agriculture, benefiting both producers and consumers.