Work

Fine identification of noxious weeds based on close-range hyperspectral imaging and spectral–spatial features of alpine meadow plants

Year: 2025

Type: article

Abstract: The spread of noxious weeds is a key indicator of alpine meadow degradation, threatening biodiversity, pasture productivity, and ecosystem stability. Accurate identification of these invasive species remains challenging due to the low stature, dense growth, and spatial heterogeneity of alpine vegetation, which limits conventional remote sensing in detecting and differentiating noxious weeds at fine spatial scales. This study proposes a close-range hyperspectral remote sensing approach that integrates both spectral and spatial features to enhance the identification of noxious weeds. Multiple features were extracted, including gradient information, textural metrics, narrow-band spectral indices (e.g., NGNDVI, NPRI), and first-order spectral derivatives, to improve the separability between noxious weeds and other plant species. Mahalanobis distance was used to select sensitive spectral bands, and four classifiers-LinearSVC, Random Forest, LightGBM, and Logistic Regression-were evaluated for classification performance. The results demonstrated that: (1) First-order derivatives, spectral indices, and texture features contributed 27.07%, 18.36%, and 11.66%, respectively, to classification performance; (2) Compared to using only spectral parameters, combining spatial and spectral features significantly improved accuracy, with the four classifiers’ Kappa coefficients increasing by 6.89%, 1.76%, 1.47%, and 4.55%; and (3) Using the optimal feature combination, overall classification accuracy exceeded 92%, with the highest Kappa coefficient reaching 0.95. These results indicate that integrating spectral and spatial features from close-range hyperspectral imaging can significantly improve the accuracy of noxious weed detection. The proposed method offers a reliable and practical approach for fine-scale vegetation monitoring and may contribute to early intervention and effective management of invasive species in fragile alpine ecosystems. • A hyperspectral method integrates spectral–spatial features to identify noxious weed. • NGNDVI, NPRI, and MEA are key variables for noxious weeds identification. • Spatial features improved noxious weeds identification: Kappa ↑ 6.89%, OA ↑ 6.16%. • Close-range hyperspectral imaging identifies noxious weeds: 95.75% OA, 95.14% Kappa.

Source: Ecological Informatics

DOI: 10.1016/j.ecoinf.2025.103430

Authors: Lei Guan, Enqin Liu, Man Yang, Bing Gao

Institution: Chengdu University of Technology

Language: English

Topic: Remote-Sensing Image Classification

Subfield: Media Technology

Field: Engineering

Domain: Physical Sciences

SDG: Life on land

Open Access status: gold

Estimated APC paid: $2,770