Skip to main content
SLU publication database (SLUpub)

Conference abstract2022Peer reviewed

Early detection of nematode-induced wilt using hyperspectral drone images

Li, Niwen; Huo, Langning; Zhang, Xiaoli

Abstract

Pine wilt disease (PWD) is a devastating pine disease caused by pine wood nematodes. It is widely distributed in North America, East Asia, and Europe, causing huge damage to production forests. The primary control measure is to cut down and remove the infected trees from forest as soon as possible to avoid the spread, which demands early identification to prevent outbreaks. This study explored the potential of hyperspectral drone images for the early detection of PWD.

The hyperspectral drone images were obtained with 0.11 m resolution and wavelength from 400 to 1000 nm, covering from red band to near-infrared (NIR). We delineated the crowns by visual interpretation and inventoried the infestation stages in the field, obtaining images of 50 healthy trees and 50 early-infected trees as samples. We extracted the spectral signatures and smoothed them with the Savitzky-Golay filter to remove the noise of hyperspectral data. We first calculated the separability of the individual bands to distinguish between healthy and early infected samples, and determined the sensitive bands according to the separability using linear discriminant analysis. Then we calculated vegetation indices using the sensitive bands and tested the separability for proposing vegetation indices. We then conducted the spectral derivative analysis and tested the separability of the first derivatives and the sensitive bands.

In the results, the red-edge bands showed the highest separability, followed by the green bands. Among the vegetation indices, the Difference Green Red and Green leaf index (GLI) achieved the highest identification rates of 73% and 72%, respectively. The first derivative at the red-edge band (wavelength 712 nm) showed the highest separability among all bands and achieved 77% of detection rates of early infected trees. We concluded that the red-edge bands were the most sensitive bands, and the derivatives at the red-edge bands can indicate the early infection. Vegetation indices of visible bands performed similar accuracy with the derivative of one red-edge band. For the next step, we plan to develop methods of using the derivatives and present the performance on the early identification of PWD.

Published in

Conference

ForestSAT, 2022-08-29 to 2022-09-03, Berlin, Germany