Researchers in Iran have integrated optical spectroscopic techniques with machine learning regression algorithms to inspect the quality of dried kiwifruit in the laboratory without damaging the samples.
Combining spectroscopy with machine learning to inspect samples without destruction
Wavelength selection algorithms and regression model comparison
Distinguishing between laboratory measurements and commercial industrial application
The editorial position of the study and prospects for agricultural artificial intelligence research
Frequently asked questions
Combining spectroscopy with machine learning to inspect samples without destruction
Researchers at the University of Mohaghegh Ardabili, Tarbiat Modares University, and Sari Agricultural Sciences and Natural Resources University in Iran have published a scientific method for predicting the quality of dried kiwifruit without destroying the tested sample. The study, published in “Scientific Reports” on October 24, 2026, combines visible and near-infrared light spectroscopy with machine learning-based regression models, representing a study with documented Iranian institutional and funding affiliation within the non-Arab Middle East.
The scientists measured the light spectra of approximately 250 dried fruit samples across wavelengths ranging from 470 to 980 nanometers, evaluating the system’s ability to predict vitamin C levels, soluble solids, and firmness degree. Spectroscopy measures how a physical body interacts with light; different chemical and physical properties affect these measurements, providing mathematical models with a means to estimate internal traits that typically require slower or sample-destroying laboratory analysis.
Wavelength selection algorithms and regression model comparison
A pivotal part of the experiment involved selecting the most useful wavelengths rather than passing the full light spectrum into each software model. Selection methods included the ant colony optimization algorithm, adaptive reweighted regional sampling, the genetic algorithm, and the minimum redundancy maximum relevance method, with thirty wavelengths selected from each method to train and evaluate alternative regression models.
The researchers compared partial least squares regression, random forests, support vector regression, and extreme gradient boosting. For vitamin C and soluble solids, support vector regression combined with ant colony optimization produced the strongest validation residual predictive deviations of approximately 5.52 and 5.17, respectively. As for firmness, partial least squares regression with adaptive reweighted regional sampling achieved a validation coefficient of determination of 0.91, a root mean square error of 0.43, and a residual predictive deviation of 3.50.
Distinguishing between laboratory measurements and commercial industrial application
These figures are purely laboratory validation metrics and do not constitute evidence of an industrial inspection device installed in Iranian food processing facilities; the coefficient of determination reflects the amount of variance captured within the studied data, and the favorable residual predictive deviation depends on the reference variance of the sample. Generalizing these results requires testing them on independent crops, various measurement devices, and different industrial processing conditions and measurement environments.
The primary scientific article recommends gathering larger external datasets and conducting tests under real-time operational conditions, as accurate predictions achieved under carefully controlled research conditions may decline if agricultural product varieties or monitoring sensors change, and no documented commercial production capacity, cost savings, or launch date has been disclosed.
The editorial position of the study and prospects for agricultural artificial intelligence research
This specialized scientific development ranks behind major regional policy stories in terms of editorial priority; although its connection to artificial intelligence is real and scientifically rigorous, it remains preliminary research in the field of agricultural product quality.
Accordingly, this study should not be described as an operational smart factory project or as having increased the nutritional value of kiwifruit thanks to artificial intelligence; rather, it is a laboratory methodological contribution that paves the way for the development of non-destructive inspection tools in the future.
Frequently asked questions
Question: How did Iranian researchers inspect dried kiwifruit quality without destroying it?
Answer: By combining visible and near-infrared light spectroscopy across wavelengths from 470 to 980 nanometers with machine learning regression models.
Question: What nutritional and physical properties did the algorithmic models predict?
Answer: The models evaluated the ability to predict vitamin C content, soluble solids ratio, and fruit firmness in 250 dried samples.
Question: Which algorithms achieved the best performance in predicting vitamin C and firmness?
Answer: Support vector regression with the ant colony optimization algorithm achieved the best results for vitamin C and solids, while partial least squares regression excelled in measuring firmness with a coefficient of determination of 0.91.
Question: Does the success of the study mean this technology is currently entering industrial production lines?
Answer: No, the results represent preliminary laboratory validation, and the researchers recommend testing them on broader datasets and under live industrial operating conditions prior to commercial application.
