Development and transfer of a non-destructive detection model based on visible/near-infrared full transmission spectroscopy for soluble solid content in pomelo under different integration times

文献类型： 外文期刊

第一作者： Xu, Sai

作者： Xu, Sai;He, Zhenhui;Liang, Xin;Xu, Sai;He, Zhenhui;Liang, Xin;Lu, Huazhong

作者机构：

关键词： Pomelo; VIS/NIR; Fruit quality; Non-destructive detection; Model transfer

期刊名称：LWT-FOOD SCIENCE AND TECHNOLOGY （ 影响因子：6.6； 五年影响因子：6.9 ）

ISSN： 0023-6438

年卷期： 2025 年 223 卷

页码：

收录情况： SCI

摘要： The thick skin and large size of pomelo make non-destructive internal quality detection a challenge for current fruit quality evaluation methods. Particularly in practical applications, soluble solids content (SSC), as an important indicator for measuring fruit sweetness and ripeness, is critical for its precise non-destructive detection, which plays a significant role in enhancing pomelo's market value. Moreover, under existing detection techniques, the size differences in pomelo necessitate the use of different integration times for spectral acquisition. The spectral variations caused by different integration times prevent the establishment of a unified detection model, limiting its development. Model transfer technology has been used to address model generalization issues, but previous studies have rarely considered the model failure due to inherent sample differences. Therefore, this study proposes a visible/near-infrared full-transmission spectroscopy method for non-destructive detection of pomelo soluble solids content, and uses model transfer to enable detection with the same model across different integration times. Spectra of the same batch of pomelo samples were collected with different integration times (140ms, 160ms, 180ms). Preprocessing operations for denoising and feature selection were performed, followed by data modeling and parameter optimization, with DS, PDS, and SST algorithms used for model transfer across different integration times. The experimental results showed that the combination of Standard Normal Variate transformation, Competitive Adaptive Reweighted Sampling algorithm, and Partial Least Squares Regression achieved the best precision, with a correlation coefficient R2 of 0.97 and a Root Mean Square Error (RMSE) of 0.16 on the validation set. The DS algorithm proved to be the optimal model transfer method, requiring only 20 calibration samples to achieve model transfer between different integration times, improving model adaptability and generalization ability. Therefore, the method proposed in this study enables rapid, non-destructive, and efficient detection of pomelo soluble solids content while being adaptable to different integration time scenarios, ensuring fruit quality. It can also guide post-harvest handling in the pomelo industry, enhancing market competitiveness and promoting industry development. The developed SNV + CARS + PLSR + DS technological framework also provides a reference for non-destructive detection of internal quality in other large-sized fruits, contributing to the standardization and intelligent advancement of agricultural non-destructive testing.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序