Detection of pine wood nematode infections in Chinese pine (Pinus tabuliformis) using hyperspectral drone images.

BACKGROUND

The pine wood nematode (PWN) has caused tremendous damage to pine forests in China. Accurately predicting the infestation stage of PWN is crucial for implementing appropriate management, such as chemically controlling early-infested trees and felling and removing trees in the severe stages of infestation. Unmanned aerial vehicle (UAV)-based hyperspectral technology can capture images with high spatial and spectral resolutions, facilitating more extensive coverage and enhanced detection efficiency. To date, few studies have used the correlation coefficient between full spectra and physiological traits to screen dual-band vegetation indices (VIs). Moreover, there is a lack of comprehensive comparison between the screened VIs, feature wavelengths, and full spectra using various machine learning methods to predict the infection stage of PWN.

RESULTS

We evaluated the abilities of screened VIs, feature wavelengths selected by successive projections algorithm (SPA), and full spectra in estimating PWN infection levels. Random forest (RF), artificial neural network (ANN), support vector machine (SVM), and three convolutional neural networks (CNN) were applied. Screened VIs performed the best (OA%: 76.03-80.99; Kappa: 0.68-0.74), and RF approach obtained highest classification accuracies (OA%: 72.73-80.99; Kappa: 0.63-0.74). In discriminating between healthy trees and PWN-infected trees at an early stage, RF using screened VIs outperformed other approaches (healthy trees: PA% = 76.92, UA% = 76.92; early-infested trees: PA% = 66.67, UA% = 72.00), and normalized difference spectral index (NDSI) selected by chlorophyll content was the most sensitive feature.

CONCLUSION

We propose the integration of RF with the screened VIs as a recommended approach for the early detection of PWN infections in Chinese Pine, which give reference to the management of PWN infections. © 2025 Society of Chemical Industry.