Articles
Sorting strawberries on a moving platform to predict total soluble solids (°Brix) using hyperspectral imaging
Article number
1465_61
Pages
409 – 418
Language
English
Abstract
Strawberries are one of the most grown fruits in the world, with about USD$ 4 billion sold in 2023. However, the quality of strawberries is difficult to estimate in a high throughput manner.
We profiled 400 retail strawberries using hyperspectral cameras (non-destructive) and a refractometer (destructive). The range of TSS (°Brix) was from 3.2 to 11.8. Both VNIR (400 to 1000 nm) Line Scan and NIR (900 to 1700 nm) Line Scan cameras (Headwall Photonics, Inc., Bolton, MA) were evaluated with strawberries on a perClass Mira® Stage with a plate moving at about 38 mm s‑1 (perClass BV, Delft, The Netherlands). The process for acquiring and processing the images, followed by modeling and 10-fold cross-validations was documented.
In terms of respective images, strawberries used in modeling (n=320) were different from those used for cross-validation (n=80). TSS (°Brix) of strawberries cannot be predicted by the VNIR Line Scan camera (400 to 1000 nm), with the highest per sample group cross validation R2 of 0.44 and SEP of 1.09. However, TSS (°Brix) of strawberry can be predicted by NIR Line Scan camera (900 to 1700 nm), with the highest per sample group cross validation R2 of 0.74 and SEP of 0.51. Our results demonstrate the ability to sort strawberries on a moving platform using hyperspectral imaging in terms of TSS (°Brix).
We profiled 400 retail strawberries using hyperspectral cameras (non-destructive) and a refractometer (destructive). The range of TSS (°Brix) was from 3.2 to 11.8. Both VNIR (400 to 1000 nm) Line Scan and NIR (900 to 1700 nm) Line Scan cameras (Headwall Photonics, Inc., Bolton, MA) were evaluated with strawberries on a perClass Mira® Stage with a plate moving at about 38 mm s‑1 (perClass BV, Delft, The Netherlands). The process for acquiring and processing the images, followed by modeling and 10-fold cross-validations was documented.
In terms of respective images, strawberries used in modeling (n=320) were different from those used for cross-validation (n=80). TSS (°Brix) of strawberries cannot be predicted by the VNIR Line Scan camera (400 to 1000 nm), with the highest per sample group cross validation R2 of 0.44 and SEP of 1.09. However, TSS (°Brix) of strawberry can be predicted by NIR Line Scan camera (900 to 1700 nm), with the highest per sample group cross validation R2 of 0.74 and SEP of 0.51. Our results demonstrate the ability to sort strawberries on a moving platform using hyperspectral imaging in terms of TSS (°Brix).
Publication
Authors
A. Hamidisepehr, M. Haarmeyer, J. Bell, C.M. Lee
Keywords
Fragaria × ananassa, non-destructive, spectral data, machine learning, quality control
Groups involved
- Division Vine and Berry Fruits
- Working Group Strawberry Culture and Management
- Division Sustaining Horticulture in a Changing World
- Division Horticulture for Human Health
- Division Plant Genetic Resources, Breeding and Biotechnology
- Division Greenhouse and Indoor Production Horticulture
- Division Precision Horticulture and Engineering
Online Articles (64)