Articles
Red-flesh kiwifruit inner quality scoring with a computer vision system
Article number
1395_45
Pages
343 – 350
Language
English
Abstract
Red-fleshed kiwifruits are a recent entry to the global market, and their nutraceutical properties have garnered significant consumer interest.
This produces investment opportunities for industries, which must accurately assess the internal quality of these fruit, particularly when faced with long shipping distances and high-quality standards.
Evaluating the redness of kiwifruit poses a complex challenge due to the inherent variability in colour localization, as well as the wide range of red shades and intensities within each fruit.
The current method employed to assess the colour quality of red-fleshed kiwifruit relies on visual inspections performed by experienced operators.
However, this method suffers from complexity, subjectivity, limited repeatability, and a slow evaluation process.
In this study, a computer vision system that exploits an unsupervised learning algorithm was developed to score fruit according both to red quantity and red quality.
RGB images of sliced fruit were segmented into the hue-saturation-value colour space to generate the RSmask, which was used to extract descriptors for the red quantity classifier.
Simultaneously, the RSmask was applied to produce a red-related image.
The red quality score was determined with a K-means classifier that assessed the descriptors derived from the conversion of the red-related image into the CIELAB colour space.
Consequently, 102 sample fruit were classified into 36 categories based on the combination of the red quantity and quality scores.
The results demonstrated that red colour quantity is much more predictable than colour quality due to human eye colour perception.
This produces investment opportunities for industries, which must accurately assess the internal quality of these fruit, particularly when faced with long shipping distances and high-quality standards.
Evaluating the redness of kiwifruit poses a complex challenge due to the inherent variability in colour localization, as well as the wide range of red shades and intensities within each fruit.
The current method employed to assess the colour quality of red-fleshed kiwifruit relies on visual inspections performed by experienced operators.
However, this method suffers from complexity, subjectivity, limited repeatability, and a slow evaluation process.
In this study, a computer vision system that exploits an unsupervised learning algorithm was developed to score fruit according both to red quantity and red quality.
RGB images of sliced fruit were segmented into the hue-saturation-value colour space to generate the RSmask, which was used to extract descriptors for the red quantity classifier.
Simultaneously, the RSmask was applied to produce a red-related image.
The red quality score was determined with a K-means classifier that assessed the descriptors derived from the conversion of the red-related image into the CIELAB colour space.
Consequently, 102 sample fruit were classified into 36 categories based on the combination of the red quantity and quality scores.
The results demonstrated that red colour quantity is much more predictable than colour quality due to human eye colour perception.
Authors
M. Piani, G. Bortolotti, D. Mengoli, N. Omodei, N. Raule, F. Spinelli, L. Manfrini
Keywords
artificial intelligence, colour perception, fruit quality, machine learning, postharvest, sorting
Groups involved
- Division Temperate Tree Nuts
- Division Vine and Berry Fruits
- Division Temperate Tree Fruits
- Division Precision Horticulture and Engineering
- Division Plant-Environment Interactions in Field Systems
- Division Tropical and Subtropical Fruit and Nuts
- Working Group Precision Management of Orchards and Vineyards
Online Articles (59)