Severity classification of pepper verticillium wilt from UAV multispectral imagery using a fuzzy deep neural network
Smart Agricultural Technology · 2025-12 · 12 · 101423
DOI: 10.1016/j.atech.2025.101423 ↗Kurz gefasst
Fachartikel in der Zeitschrift Smart Agricultural Technology, 2025-12. Verfasst von Lijian Lu und 5 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: National Natural Science Foundation of China, Shihezi University.
Abstract (Originalsprache)
To address the problem of low classification accuracy and limited detection efficiency of pepper verticillium wilt (VW) caused by fuzzy, uncertain information, this study integrates fuzzy logic into a deep learning framework for verticillium wilt severity (VWS) classification based on unmanned aerial vehicle (UAV) multispectral imagery. Leveraging UAV remote sensing, a multimodal feature set comprising 20 vegetation indices, together with UAV-derived spectral bands, was constructed. principal component analysis (PCA) was applied to reduce dimensionality, resulting in two datasets: a full feature set (FF, 24 features) and a selected feature set (SF, 10 features). Three machine learning models-including Random Forest (RF), Support Vector Regression (SVR), and eXtreme Gradient Boosting (XGBoost)-were optimized via grid search. In addition, two deep learning models were developed, namely the one-dimensional convolutional neural network (1DCNN) and the fuzzy-enhanced 1DCNN (fuzzy 1DCNN). The experimental results showed that SVR performed satisfactorily among the machine learning models ( F 1 s c o r e = 0.858, O A = 0.874), but the deep learning models outperformed the traditional methods overall. The fuzzy 1DCNN yielded the highest classification accuracy and robustness ( F 1 s c o r e = 0.915, O A = 0.926). Notably, it maintained superior performance even on the FF dataset, demonstrating strong resistance to noise, as well as irrelevant features. The integration of fuzzy membership functions enabled adaptive weighting of input variables, thereby enhancing fault tolerance and improving feature discrimination. Overall, the proposed fuzzy deep neural network shows strong potential for managing uncertainty in remote sensing data and offers a robust framework for the real-time monitoring and precise field management of pepper crops affected by VW.
Förderung
- National Natural Science Foundation of China
- Shihezi University
Zitieren
APA
Lu, L., Lei, J., Qin, X., Zhang, J., Wang, S., & Wang, C. (2025). Severity classification of pepper verticillium wilt from UAV multispectral imagery using a fuzzy deep neural network. Smart Agricultural Technology, 12, 101423. https://doi.org/10.1016/j.atech.2025.101423
Vancouver
Lu L, Lei J, Qin X, Zhang J, Wang S, Wang C. Severity classification of pepper verticillium wilt from UAV multispectral imagery using a fuzzy deep neural network. Smart Agricultural Technology. 2025;12:101423. doi:10.1016/j.atech.2025.101423
BibTeX
@article{Lu2025,
author = {Lu, Lijian and Lei, Jin and Qin, Xinyan and Zhang, Janglong and Wang, Shiguo and Wang, Chengfu},
title = {Severity classification of pepper verticillium wilt from UAV multispectral imagery using a fuzzy deep neural network},
journal = {Smart Agricultural Technology},
year = {2025},
volume = {12},
pages = {101423},
doi = {10.1016/j.atech.2025.101423}
}
EUVIMED