A Machine Learning Approach to Detecting Pine Wilt Disease Using Airborne Spectral Imagery

Abstract

Pine Wilt Disease is one of the most destructive pests affecting coniferous forests. After being infected by the harmful Bursaphelenchus xylophilus nematode, most trees die within one year. The complex spreading pattern of the disease and the tedious hard labor process of diagnosis involving field wood sampling followed by laboratory analysis call for alternative methods to detect and manage the infected areas. Remote sensing comes naturally into play owing to the possibility of covering relatively large areas and the ability to discriminate healthy from sick trees based on spectral characteristics. This paper presents the development of machine learning classification algorithms for the detection of Pine Wilt Disease in Pinus pinaster, performed in the framework of the European Commission’s Horizon 2020 project “Operational Forest Monitoring using Copernicus and UAV Hyperspectral Data” (FOCUS) in two provinces of central Portugal. Five flight campaigns have been carried out in two consecutive years in order to capture a multitemporal variation of disease distribution. Classification algorithms based on a Random Forest approach were separately designed for the acquired very-high-resolution multispectral and hyperspectral data, respectively. Both algorithms achieved overall accuracies higher than 0.91 in test data. Furthermore, our study shows that the early detection of decaying trees is feasible, even before symptoms are visible in the field.

Detection of Longhorned Borer Attack and Assessment in Eucalyptus Plantations Using UAV Imagery

“…This precision is explained by the great distance between reflectance values and the differences between the two classes. Iordache et al [28] applied RF classifier on the classification of Pinus pinaster canopy types (infected, suspicious, and healthy) affected by pine wild and obtained an overall accuracy of 95%. Pourazar et al [27], obtain as overall accuracy 95.58% using five spectral bands and five indices to detect dead and diseased trees.…”

“…In order to classify tree canopies into two different classes, healthy and dead trees, supervised machine-learning (ML) classification was used. Among the different object-oriented ML classifiers, the RF algorithm was applied as its performance is one of the most accurate ML algorithms when supervised classification for GEOBIA is conducted [27,28,68,73]. Presented by Breiman [74], RF is an automatic ensemble method based on decision trees where each tree depends on a collection of random variables [75].…”

“…In recent years the use of unmanned aerial vehicle (UAV) platforms has become widely employed for pests and disease detection and monitoring [12,[22][23][24][25][26][27][28]. The main attributes are very high resolution, suitability for multitemporal analysis, lower operational costs compared with airplanes and satellites, independence of cloud cover, and the ability to operate in specific phenological phases of plants or pest/disease outbreaks [12].…”

“…For comparisons, another classification used was the OBIA with a random forest algorithm and an overall accuracy of 93%. Finally, Iordache et al [28] studied pine wild disease in Portugal using OBIA and machine-learning random forest algorithm for multispectral and hyperspectral imagery. The overall accuracy of both classifications was 95% and 91%, respectively.…”

Spectroscopic detection of forest diseases: a review (1970–2020)

“…Recently, Iordache et al (2020) proposed a machine learning algorithms detection of PWD in Pinus pinaster Aiton (maritime pine) from multi-(MicaSense Red-Edge M multispectral camera, MicaSense Inc.; five spectral bands centered at 475, 560, 668, 717, and 840 nm) and hyperspectral data (MicroHyperSpec A-series hyperspectral scanner, Headwall Photonics; 380−1100 nm) acquired over two regions of central Portugal affected by PWD. Classification schemes for both multi-and hyperspectral imagery (spatial resolution: 5 and 10 cm, respectively) used sets of 13 selected VSI, and were developed by random forest approaches.…”