# Nondestructive Evaluation of Soluble Solid Content of Cucumbers Based on VIS–NIR and SWIR Hyperspectral Images

## TL;DR

This study uses hyperspectral imaging to nondestructively measure the quality of cucumbers, showing that visible–near infrared (VIS–NIR) is more cost-effective for large-scale use.

## Contribution

The study introduces optimized PLSR models for nondestructive SSC detection using VIS–NIR and SWIR hyperspectral imaging in cucumbers.

## Key findings

- VIS–NIR achieved an R²p of 0.827 with RMSEP of 0.176 for SSC prediction.
- SWIR achieved an R²p of 0.818 with RMSEP of 0.177 for SSC prediction.
- VIS–NIR is more cost-effective than SWIR for online monitoring in cucumber sorting.

## Abstract

Soluble solid content (SSC) is a key indicator for evaluating cucumber quality, directly influencing its commercial value. In China's cucumber sorting factories, SSC is typically assessed using random sampling and destructive methods, which are unsuitable for large‐scale and continuous detection. Therefore, this study employs hyperspectral imaging technology to evaluate the capability of visible–near infrared (VIS–NIR) and shortwave infrared (SWIR) spectroscopy for nondestructive SSC detection. In the experiment, hyperspectral data of cucumbers at different growth stages were collected in the VIS–NIR and SWIR. Using a partial least squares regression (PLSR) model, SSC prediction performance was compared across three spectral preprocessing methods and three sensitive wavelength selection methods. The optimal prediction models for SSC in the VIS–NIR and SWIR spectral ranges were established. The results showed that the optimal model in the VIS–NIR was the savitzky–golay smoothing (SG)‐fullwave‐PLSR model, with an R2
p of 0.827, an RMSEP of 0.176, and an RPD of 2.403. In the SWIR, the optimal model was the multiplicative scatter correction (MSC)‐competitive adaptive reweighted sampling (CARS)‐PLSR, with an R
2
p of 0.818, an RMSEP of 0.177, and an RPD of 2.344. The study demonstrates that hyperspectral imaging in both VIS–NIR and SWIR can be applied for nondestructive SSC detection in cucumber sorting factories. Considering both prediction accuracy and cost, VIS–NIR is more suitable for online monitoring of cucumber SSC.

This study employs hyperspectral imaging (VIS–NIR/SWIR) for nondestructive soluble solid content (SSC) detection in cucumbers. Optimized PLSR models with preprocessing and sensitive wavelength selection achieved R
2
p of 0.827 (VIS–NIR) and 0.818 (SWIR). Results demonstrate that VIS–NIR and SWIR are viable for cucumber SSC detection, with VIS–NIR being more cost‐effective for online monitoring in sorting facilities.