Λεπτομέρειες βιβλιογραφικής εγγραφής
| Τίτλος: |
Damage Identification of Cross‐Variety Kiwifruit Based on Hyperspectral Imaging Combined With Transfer Learning and Semi‐Supervised Self‐Training. |
| Συγγραφείς: |
Sun, Jun1 (AUTHOR), Cong, Sunli1 (AUTHOR) 2112407120@stmail.ujs.edu.cn, Ma, Pengpeng1 (AUTHOR), Wu, Xiaohong1 (AUTHOR), Dai, Chunxia1 (AUTHOR), Zhou, Xin1 (AUTHOR) |
| Πηγή: |
Journal of Food Process Engineering. Sep2026, Vol. 49 Issue 9, p1-14. 14p. |
| Θεματικοί όροι: |
*Hyperspectral imaging systems, *Nondestructive testing, Autoencoders, Transfer of training, Support vector machines |
| Περίληψη: |
Kiwifruit is prone to mechanical damage during storage and transportation after harvest, and the early‐stage damage is hidden. Traditional manual detection methods have problems such as low efficiency and strong subjectivity. In response to the problem of decreased generalization ability of models caused by differences in spectral distribution between different varieties, this study took "Yate" and "Xuxiang" kiwifruit as research objects, and proposed a cross‐variety damage identification method that combines hyperspectral imaging with transfer learning and semi‐supervised self‐training (ST). The hyperspectral images of kiwifruit samples in different damage stages were collected, and the extracted spectral data were preprocessed. Afterwards, the support vector machine (SVM) model based on single variety, mixed varieties and the SVM model based on autoencoder (AE) feature transfer (AE‐SVM) were constructed, respectively. The results show that the AE‐SVM model extracts 16‐dimensional latent features by means of "source domain (Yate) training‐target domain (Xuxiang) projection," so that the target domain samples are expressed in the unified latent space, yielding an average test‐set accuracy of 93.18% and an average test‐set F1 score of 93.15%, which proves that AE feature transfer can effectively alleviate the differences in cross‐variety spectral distribution. On this basis, the ST mechanism was further introduced to expand the training set by screening high‐confidence pseudo‐labeled samples, so as to improve the coverage ability of the model to the feature space of the target domain. The classification average accuracy and F1 score of the final model (ST‐AE‐SVM) on the "Xuxiang" independent test set were improved to 95.45% and 95.42%, respectively. Therefore, hyperspectral imaging combined with AE feature transfer and semi‐supervised ST can effectively improve the damage identification effect of cross‐variety kiwifruit, and provide reference for intelligent sorting and non‐destructive quality testing of postharvest fruits. [ABSTRACT FROM AUTHOR] |
|
Copyright of Journal of Food Process Engineering is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) |
| Βάση Δεδομένων: |
Business Source Index |