Small unmanned aerial vehicle (UAV) based remote sensing is a rapidly evolving technology. Novel sensors and methods are entering the market, offering completely new possibilities to carry out remote sensing tasks. Three-dimensional (3D) hyperspectral remote sensing is a novel and powerful technology that has recently become available to small UAVs. This study investigated the performance of UAV-based photogrammetry and hyperspectral imaging in individual tree detection and tree species classification in boreal forests. Eleven test sites with 4151 reference trees representing various tree species and developmental stages were collected in June 2014 using a UAV remote sensing system equipped with a frame format hyperspectral camera and an RGB camera in highly variable weather conditions. Dense point clouds were measured photogrammetrically by automatic image matching using high resolution RGB images with a 5 cm point interval. Spectral features were obtained from the hyperspectral image blocks, the large radiometric variation of which was compensated for by using a novel approach based on radiometric block adjustment with the support of in-flight irradiance observations. Spectral and 3D point cloud features were used in the classification experiment with various classifiers. The best results were obtained with Random Forest and Multilayer Perceptron (MLP) which both gave 95% overall accuracies and an F-score of 0.93. Accuracy of individual tree identification from the photogrammetric point clouds varied between 40% and 95%, depending on the characteristics of the area. Challenges in reference measurements might also have reduced these numbers. Results were promising, indicating that hyperspectral 3D remote sensing was operational from a UAV platform even in very difficult conditions. These novel methods are expected to provide a powerful tool for automating various environmental close-range remote sensing tasks in the very near future.

Recognition of tree species and geospatial information on tree species composition is essential for forest management. In this study, tree species recognition was examined using hyperspectral imagery from visible to near-infrared (VNIR) and short-wave infrared (SWIR) camera sensors in combination with a 3D photogrammetric canopy surface model based on RGB camera stereo-imagery. An arboretum with a diverse selection of 26 tree species from 14 genera was used as a test area. Aerial hyperspectral imagery and high spatial resolution photogrammetric color imagery were acquired from the test area using unmanned aerial vehicle (UAV) borne sensors. Hyperspectral imagery was processed to calibrated reflectance mosaics and was tested along with the mosaics based on original image digital number values (DN). Two alternative classifiers, a k nearest neighbor method (k-nn), combined with a genetic algorithm and a random forest method, were tested for predicting the tree species and genus, as well as for selecting an optimal set of remote sensing features for this task. The combination of VNIR, SWIR, and 3D features performed better than any of the data sets individually. Furthermore, the calibrated reflectance values performed better compared to uncorrected DN values. These trends were similar with both tested classifiers. Of the classifiers, the k-nn combined with the genetic algorithm provided consistently better results than the random forest algorithm. The best result was thus achieved using calibrated reflectance features from VNIR and SWIR imagery together with 3D point cloud features; the proportion of correctly-classified trees was 0.823 for tree species and 0.869 for tree genus.

In this study we compared the accuracy of low-pulse airborne laser scanning (ALS) data, multi-temporal high-resolution noninterferometric TerraSAR-X radar data and a combined feature set derived from these data in the estimation of forest variables at plot level. The TerraSAR-X data set consisted of seven dual-polarized (HH/HV or VH/VV) Stripmap mode images from all seasons of the year. We were especially interested in distinguishing between the tree species. The dependent variables estimated included mean volume, basal area, mean height, mean diameter and tree species-specific mean volumes. Selection of best possible feature set was based on a genetic algorithm (GA). The nonparametric k-nearest neighbour (k-NN) algorithm was applied to the estimation. The research material consisted of 124 circular plots measured at tree level and located in the vicinity of Espoo, Finland. There are large variations in the elevation and forest structure in the study area, making it demanding for image interpretation. The best feature set contained 12 features, nine of them originating from the ALS data and three from the TerraSAR-X data. The relative RMSEs for the best performing feature set were 34.7% (mean volume), 28.1% (basal area), 14.3% (mean height), 21.4% (mean diameter), 99.9% (mean volume of Scots pine), 61.6% (mean volume of Norway spruce) and 91.6% (mean volume of deciduous tree species). The combined feature set outperformed an ALS-based feature set marginally; in fact, the latter was better in the case of species-specific volumes. Features from TerraSAR-X alone performed poorly. However, due to favorable temporal resolution, satellite-borne radar imaging is a promising data source for updating large-area forest inventories based on low-pulse ALS.