Key Points
- Removing pomelo peels before visible-near-infrared (vis-NIR) spectroscopy significantly improves the accuracy of soluble solids content (SSC) predictions.
- The study's peeled fruit model using partial least squares regression achieved higher predictive accuracy (R²P = 0.88, RPD = 2.57) than models using intact fruit.
- Researchers also applied 1D convolutional neural networks (1D-CNN), which matched traditional methods while offering greater automation and scalability for postharvest fruit quality assessment.
A recent study published in the journal Food Control investigated a new method that could potentially improve the accuracy of soluble solids content in pomelos (1). This study, which was led by a team of researchers from Jiangsu University and Zhejiang University, demonstrated the value of removing the thick peel of pomelos before using visible-near-infrared (vis-NIR) spectroscopy to assess their quality.
Pomelos are a type of citrus fruit that is popular in several Asian countries, especially China (2). It is the largest member of the citrus family (2). Apart from their large size, pomelos are known for a thick outer rind, which poses a challenge for traditional detection methods. Because pomelo peels are much thicker, traditional vis-NIR spectroscopy often encounters spectral interference when evaluating SSC for thicker-skinned fruits (1). This interference not only affects the visible spectral band but also alters the overall spectral absorption strength, leading to diminished predictive reliability (1).
In their study, the research team investigated how to better assess the quality of pomelos by improving on traditional methods. They accomplished this by suggesting a new postharvest processing model, which evaluates the fruit after peeling without altering its marketability or edibility. The main goal was to verify the extent to which the peel interferes with SSC detection, and to develop a more precise method for evaluating the internal quality of pomelos (1).
To investigate this, the researchers developed a specialized online detection system capable of capturing transmission spectra from both intact and peeled pomelos. They then performed a comparative analysis of the spectral characteristics and constructed several predictive models using both conventional partial least squares regression (PLSR) and modern deep learning techniques (1).
What Spectral Preprocessing Methods Were Used in the Study?
The spectral preprocessing methods used in this study were standard normal variate (SNV) and second-order detrending. These methods were applied to improve the performance of their model. Feature selection was also optimized using changeable-size moving window (CSMW) algorithms (1). Among the various models, the PLSR model built on spectra from peeled pomelos showed the highest predictive accuracy, with a determination coefficient of prediction (R²P) of 0.88, a root mean square error of prediction (RMSEP) of 0.294%, and a residual predictive deviation (RPD) of 2.57 (1). In contrast, the model based on intact pomelo spectra only achieved an RPD value of 1.91 (1).
These findings highlight the significant negative impact of the peel on wavelength variable selection and overall model prediction. Removing the peel before spectral acquisition allowed for much clearer signals and more precise evaluation of SSC (1).
What Intelligent Modeling Did the Researchers Use in Their Study?
To interpret the spectral data sets, the researchers used one-dimensional convolutional neural network (1D-CNN). They used 1D-CNN because the model did not require manual preprocessing or wavelength selection; this means that it automatically extracts relevant features during training (1). The researchers found that the CNN model not only matched the performance of the best PLSR models, but it also offered greater potential for scalability and automation in real-world applications.
By demonstrating that peeling enhances SSC detection accuracy without compromising the fruit's integrity or consumer appeal, the researchers showed a novel postharvest processing strategy that works, and that involves grading after peeling.
Large-scale citrus production could benefit from this approach, since quality sorting needs to be reliable and efficient. Particularly for fruits with thick or variable peels, the researchers demonstrated in this study how vis-NIR, when integrated with intelligent modeling, can improve accuracy and reliable of citrus fruit quality detection (1).
References
- Wang, C.; Luo, X.; Guo, Z.; et al. Influence of the Peel on Online Detecting Soluble Solids Content of Pomelo Using Vis-NIR Spectroscopy Coupled with Chemometric Analysis. Food Cont. 2025, 167, 110777. DOI: 10.1016/j.foodcont.2024.110777
- Riske, H. What is a Pomelo? Everything to Know About This Citrus. Better Homes & Gardens. Available at: https://www.bhg.com/what-is-a-pomelo-7095474#:~:text=What%20Are%20Pomelos?,is%20usually%20removed%20before%20eating. (accessed 2025-06-06).
