Species Classification of Automatically Delineated Regenerating Conifer Crowns Using RGB and Near-Infrared UAV Imagery

Andrew J Chadwick,Lee A Martens,Christopher W Bater,Barry White,Nicholas C Coops

doi:10.1109/lgrs.2021.3123552

Abstract

Unmanned aerial vehicles (UAVs) and deep learning are important tools at the forefront of automated forest monitoring research, where classification of individual tree species is a critical forest management goal. Near-infrared (NIR) information provided by specialized UAV sensors may improve classification accuracy at the cost of added operational complexity; however, this potential for improvement is context-dependent and, therefore, may not be necessary. We assessed the performance of conventional red-green-blue (RGB) versus NIR imagery when classifying regenerating lodgepole pine and white spruce crowns automatically delineated by a trained deep learning algorithm. Models trained on NIR imagery slightly outperformed those trained on RGB imagery. Models trained on spectral bands outperformed those trained on spectral indices. The minor difference in performance between the two sets of imagery showed that accurate classification of lodgepole pine and white spruce can be carried-out using conventional RGB imagery.

Highlights

The simultaneous proliferation of Unmanned Aerial Vehicles (UAV) remote sensing and deep learning techniques has accelerated research into high spatial resolution object detection [1]
We developed a pipeline for the automatic delineation of regenerating lodgepole pine (Pinus contorta Dougl. ex Loud. var latifolia Engelm., Pl) and white spruce (Picea glauca (Moench) Voss, Sw) crowns using Mask RCNN and fine spatial resolution RGB imagery [7]
We assessed the value of NIR information versus conventional RGB information for the classification of lodgepole pine (Pl) and white spruce (Sw) crowns, which were automatically delineated using a trained instance of Mask RCNN

Summary

Introduction

The simultaneous proliferation of UAV remote sensing and deep learning techniques has accelerated research into high spatial resolution object detection [1]. Instance segmentation on high spatial resolution imagery via advanced deep learning algorithms is likely an effective avenue for automated analyses of regenerating stands at the individual tree crown (ITC) scale, as treessituated in open canopies can be relatively discerned from backgrounds comprised of exposed soil or non-tree vegetation [4], [5]. The inclusion of spectral bands in the near infrared region of the electromagnetic spectrum, in addition to conventional visible bands, on UAV camera systems offers potential to increase discrimination and accuracies of species classification [8], the usefulness of a given spectral band is dependent on the species being classified. Immitzer et al [10] found SWIR (short-wave infrared) and red-edge bands from Sentinel 2 to be most important for the separation of deciduous species, whereas the red band was most important for the separation of coniferous species

Methods

Discussion

Conclusion