Object-Based Tree Species Classification Using Airborne Hyperspectral Images and LiDAR Data

Identification of tree species based on the fusion of UAV hyperspectral image and LiDAR data in a coniferous and broad-leaved mixed forest in Northeast China

Rapid and accurate identification of tree species via remote sensing technology has become one of the important means for forest inventory. This paper is to develop an accurate tree species identification framework that integrates unmanned airborne vehicle (UAV)-based hyperspectral image and light detection and ranging (LiDAR) data under the complex condition of natural coniferous and broad-leaved mixed forests. First, the UAV-based hyperspectral image and LiDAR data were obtained from a natural coniferous and broad-leaved mixed forest in the Maoer Mountain area of Northeast China. The preprocessed LiDAR data was segmented using a distance-based point cloud clustering algorithm to obtain the point cloud of individual trees; the hyperspectral image was segmented using the projection outlines of individual tree point clouds to obtain the hyperspectral data of individual trees. Then, different hyperspectral and LiDAR features were extracted, respectively, and the importance of the features was analyzed by a random forest (RF) algorithm in order to select appropriate features for the single-source and multi-source data. Finally, tree species identification in the study area were conducted by using a support vector machine (SVM) algorithm together with hyperspectral features, LiDAR features and fused features, respectively. Results showed that the total accuracy for individual tree segmentation was 84.62%, and the fused features achieved the best accuracy for identification of the tree species (total accuracy = 89.20%), followed by the hyperspectral features (total accuracy = 86.08%) and LiDAR features (total accuracy = 76.42%). The optimal features for tree species identification based on fusion of the hyperspectral and LiDAR data included the vegetation indices that were sensitive to the chlorophyll, anthocyanin and carotene contents in the leaves, the partial components of the transformed independent component analysis (ICA), minimum noise fraction (MNF) and principal component analysis (PCA), and the intensity features of the LiDAR echo, respectively. It was concluded that the framework developed in this study was effective in tree species identification under the complex conditions of natural coniferous and broad-leaved mixed forest and the fusion of UAV-based hyperspectral image and LiDAR data can achieve enhanced accuracy compared the single-source UAV-based remote sensing data.

Tropical Species Classification with Structural Traits Using Handheld Laser Scanning Data

Information about tree species plays a pivotal role in sustainable forest management. Light detection and ranging (LiDAR) technology has demonstrated its potential to obtain species information using the structural features of trees. Several studies have explored the structural properties of boreal or temperate trees from terrestrial laser scanning (TLS) data and applied them to species classification, but the study of structural properties of tropical trees for species classification is rare. Compared to conventional static TLS, handheld laser scanning (HLS) is able to effectively capture point clouds of an individual tree with flexible movability. Therefore, in this study, we characterized the structural features of tropical species from HLS data as 23 LiDAR structural parameters, involving 6 branch, 11 crown and 6 entire tree parameters, and used these parameters to classify the species via 5 machine-learning (ML) models, respectively. The performance of each parameter was further evaluated and compared. Classification results showed that the employed parameters can achieve a classification accuracy of 84.09% using the support vector machine with a polynomial kernel. The evaluation of parameters indicated that it is insufficient to classify four species with only one and two parameters, but ten parameters were recommended in order to achieve satisfactory accuracy. The combination of different types of parameters, such as branch and crown parameters, can significantly improve classification accuracy. Finally, five sets of optimal parameters were suggested according to their importance and performance. This study also showed that the time- and cost-efficient HLS instrument could be a promising tool for tree-structure-related studies, such as structural parameter estimation, species classification, forest inventory, as well as sustainable tree management.

Classification of Tree Species in Different Seasons and Regions Based on Leaf Hyperspectral Images

This paper aims to establish a tree species identification model suitable for different seasons and regions based on leaf hyperspectral images, and to mine a more effective hyperspectral identification algorithm. Firstly, the reflectance spectra of leaves in different seasons and regions were analyzed. Then, to solve the problem that 0-element in sparse random (SR) coding matrices affects the classification performance of error-correcting output codes (ECOC), two versions of supervision-mechanism-based ECOC algorithms, namely SM-ECOC-V1 and SM-ECOC-V2, were proposed in this paper. In addition, the performance of the proposed algorithms was compared with that of six traditional algorithms based on all bands and feature bands. The experiment results show that seasonal and regional changes have an effect on the reflectance spectra of leaves, especially in the near-infrared region of 760–1000 nm. When the spectral information of different seasons and different regions is added into the identification model, tree species can be effectively classified. SM-ECOC-V2 achieves the best classification performance based on both all bands and feature bands. Furthermore, both SM-ECOC-V1 and SM-ECOC-V2 outperform the ECOC method under SR coding strategy, indicating the proposed methods can effectively avoid the influence of 0-element in SR coding matrix on classification performance.

Classifying Crop Types Using Two Generations of Hyperspectral Sensors (Hyperion and DESIS) with Machine Learning on the Cloud

Advances in spaceborne hyperspectral (HS) remote sensing, cloud-computing, and machine learning can help measure, model, map and monitor agricultural crops to address global food and water security issues, such as by providing accurate estimates of crop area and yield to model agricultural productivity. Leveraging these advances, we used the Earth Observing-1 (EO-1) Hyperion historical archive and the new generation DLR Earth Sensing Imaging Spectrometer (DESIS) data to evaluate the performance of hyperspectral narrowbands in classifying major agricultural crops of the U.S. with machine learning (ML) on Google Earth Engine (GEE). EO-1 Hyperion images from the 2010–2013 growing seasons and DESIS images from the 2019 growing season were used to classify three world crops (corn, soybean, and winter wheat) along with other crops and non-crops near Ponca City, Oklahoma, USA. The supervised classification algorithms: Random Forest (RF), Support Vector Machine (SVM), and Naive Bayes (NB), and the unsupervised clustering algorithm WekaXMeans (WXM) were run using selected optimal Hyperion and DESIS HS narrowbands (HNBs). RF and SVM returned the highest overall producer’s, and user’s accuracies, with the performances of NB and WXM being substantially lower. The best accuracies were achieved with two or three images throughout the growing season, especially a combination of an earlier month (June or July) and a later month (August or September). The narrow 2.55 nm bandwidth of DESIS provided numerous spectral features along the 400–1000 nm spectral range relative to smoother Hyperion spectral signatures with 10 nm bandwidth in the 400–2500 nm spectral range. Out of 235 DESIS HNBs, 29 were deemed optimal for agricultural study. Advances in ML and cloud-computing can greatly facilitate HS data analysis, especially as more HS datasets, tools, and algorithms become available on the Cloud.

Analysis of Regional Distribution of Tree Species Using Multi-Seasonal Sentinel-1&2 Imagery within Google Earth Engine

Accurate information on tree species is in high demand for forestry management and further investigations on biodiversity and environmental monitoring. Over regional or large areas, distinguishing tree species at high resolutions faces the challenges of a lack of representative features and computational power. A novel methodology was proposed to delineate the explicit spatial distribution of six dominant tree species (Pinus tabulaeformis, Quercus mongolia, Betula spp., Populus spp., Larix spp., and Armeniaca sibirica) and one residual class at 10 m resolution. Their spatial patterns were analyzed over an area covering over 90,000 km2 using the analysis-ready large volume of multisensor imagery within the Google Earth engine (GEE) platform afterwards. Random forest algorithm built into GEE was used together with the 20th and 80th percentiles of multitemporal features extracted from Sentinel-1/2, and topographic features. The composition of tree species in natural forests and plantations at the city and county-level were performed in detail afterwards. The classification achieved a reliable accuracy (77.5% overall accuracy, 0.71 kappa), and the spatial distribution revealed that plantations (Pinus tabulaeformis, Populus spp., Larix spp., and Armeniaca sibirica) outnumber natural forests (Quercus mongolia and Betula spp.) by 6% and were mainly concentrated in the northern and southern regions. Arhorchin had the largest forest area of over 4500 km2, while Hexingten and Aohan ranked first in natural forest and plantation area. Additionally, the class proportion of the number of tree species in Karqin and Ningcheng was more balanced. We suggest focusing more on the suitable areas modeling for tree species using species’ distribution models and environmental factors based on the classification results rather than field survey plots in further studies.

Feature-Level Fusion between Gaofen-5 and Sentinel-1A Data for Tea Plantation Mapping

The accurate mapping of tea plantations is significant for government decision-making and environmental protection of tea-producing regions. Hyperspectral and Synthetic Aperture Radar (SAR) data have recently been widely used in land cover classification, but effective integration of these data for tea plantation mapping requires further study. This study developed a new feature-level image fusion method called LPPSubFus that combines locality preserving projection and subspace fusion (SubFus) to map tea plantations. Based on hyperspectral and SAR data, we first extracted spectral indexes, textures, and backscattering information. Second, this study applied LPPSubFus to tea plantation mapping with different classification algorithms. Finally, we compared the performance of LPPSubFus, SubFus, and pixel-level image fusion in tea plantation mapping. Feature-level image fusion performed better than pixel-level image fusion. An improvement of about 3% was achieved using feature-level image fusion compared to hyperspectral data alone. Regarding feature-level image fusion, LPPSubFus improved the overall accuracy by more than 3% compared to SubFus. In particular, LPPSubFus using neural network algorithms achieved the highest overall accuracy (95%) and over 90% producer and user accuracy for tea plantations and forests. In addition, LPPSubFus was more compatible with different classification algorithms than SubFus. Based on these findings, it is concluded that LPPSubFus has better and more stable performance in tea plantation mapping than pixel-level image fusion and SubFus. This study demonstrates the potential of integrating hyperspectral and SAR data via LPPSubFus for mapping tea plantations. Our work offers a promising tea plantation mapping method and contributes to the understanding of hyperspectral and SAR data fusion.

Improved Prototypical Network Model for Forest Species Classification in Complex Stand

Deep learning has become an effective method for hyperspectral image classification. However, the high band correlation and data volume associated with airborne hyperspectral images, and the insufficiency of training samples, present challenges to the application of deep learning in airborne image classification. Prototypical networks are practical deep learning networks that have demonstrated effectiveness in handling small-sample classification. In this study, an improved prototypical network is proposed (by adding L2 regularization to the convolutional layer and dropout to the maximum pooling layer) to address the problem of overfitting in small-sample classification. The proposed network has an optimal sample window for classification, and the window size is related to the area and distribution of the study area. After performing dimensionality reduction using principal component analysis, the time required for training using hyperspectral images shortened significantly, and the test accuracy increased drastically. Furthermore, when the size of the sample window was 27 × 27 after dimensionality reduction, the overall accuracy of forest species classification was 98.53%, and the Kappa coefficient was 0.9838. Therefore, by using an improved prototypical network with a sample window of an appropriate size, the network yielded desirable classification results, thereby demonstrating its suitability for the fine classification and mapping of tree species.

Detecting European Aspen (Populus tremula L.) in Boreal Forests Using Airborne Hyperspectral and Airborne Laser Scanning Data

Sustainable forest management increasingly highlights the maintenance of biological diversity and requires up-to-date information on the occurrence and distribution of key ecological features in forest environments. European aspen (Populus tremula L.) is one key feature in boreal forests contributing significantly to the biological diversity of boreal forest landscapes. However, due to their sparse and scattered occurrence in northern Europe, the explicit spatial data on aspen remain scarce and incomprehensive, which hampers biodiversity management and conservation efforts. Our objective was to study tree-level discrimination of aspen from other common species in northern boreal forests using airborne high-resolution hyperspectral and airborne laser scanning (ALS) data. The study contained multiple spatial analyses: First, we assessed the role of different spectral wavelengths (455–2500 nm), principal component analysis, and vegetation indices (VI) in tree species classification using two machine learning classifiers—support vector machine (SVM) and random forest (RF). Second, we tested the effect of feature selection for best classification accuracy achievable and third, we identified the most important spectral features to discriminate aspen from the other common tree species. SVM outperformed the RF model, resulting in the highest overall accuracy (OA) of 84% and Kappa value (0.74). The used feature set affected SVM performance little, but for RF, principal component analysis was the best. The most important common VI for deciduous trees contained Conifer Index (CI), Cellulose Absorption Index (CAI), Plant Stress Index 3 (PSI3), and Vogelmann Index 1 (VOG1), whereas Green Ratio (GR), Red Edge Inflection Point (REIP), and Red Well Position (RWP) were specific for aspen. Normalized Difference Red Edge Index (NDRE) and Modified Normalized Difference Index (MND705) were important for coniferous trees. The most important wavelengths for discriminating aspen from other species included reflectance bands of red edge range (724–727 nm) and shortwave infrared (1520–1564 nm and 1684–1706 nm). The highest classification accuracy of 92% (F1-score) for aspen was achieved using the SVM model with mean reflectance values combined with VI, which provides a possibility to produce a spatially explicit map of aspen occurrence that can contribute to biodiversity management and conservation efforts in boreal forests.

Deep Learning for Land Cover Classification Using Only a Few Bands

There is an emerging interest in using hyperspectral data for land cover classification. The motivation behind using hyperspectral data is the notion that increasing the number of narrowband spectral channels would provide richer spectral information and thus help improve the land cover classification performance. Although hyperspectral data with hundreds of channels provide detailed spectral signatures, the curse of dimensionality might lead to degradation in the land cover classification performance. Moreover, in some practical applications, hyperspectral data may not be available due to cost, data storage, or bandwidth issues, and RGB and near infrared (NIR) could be the only image bands available for land cover classification. Light detection and ranging (LiDAR) data is another type of data to assist land cover classification especially if the land covers of interest have different heights. In this paper, we examined the performance of two Convolutional Neural Network (CNN)-based deep learning algorithms for land cover classification using only four bands (RGB+NIR) and five bands (RGB+NIR+LiDAR), where these limited number of image bands were augmented using Extended Multi-attribute Profiles (EMAP). The deep learning algorithms were applied to a well-known dataset used in the 2013 IEEE Geoscience and Remote Sensing Society (GRSS) Data Fusion Contest. With EMAP augmentation, the two deep learning algorithms were observed to achieve better land cover classification performance using only four bands as compared to that using all 144 hyperspectral bands.

Integrating National Ecological Observatory Network (NEON) Airborne Remote Sensing and In-Situ Data for Optimal Tree Species Classification

Accurately mapping tree species composition and diversity is a critical step towards spatially explicit and species-specific ecological understanding. The National Ecological Observatory Network (NEON) is a valuable source of open ecological data across the United States. Freely available NEON data include in-situ measurements of individual trees, including stem locations, species, and crown diameter, along with the NEON Airborne Observation Platform (AOP) airborne remote sensing imagery, including hyperspectral, multispectral, and light detection and ranging (LiDAR) data products. An important aspect of predicting species using remote sensing data is creating high-quality training sets for optimal classification purposes. Ultimately, manually creating training data is an expensive and time-consuming task that relies on human analyst decisions and may require external data sets or information. We combine in-situ and airborne remote sensing NEON data to evaluate the impact of automated training set preparation and a novel data preprocessing workflow on classifying the four dominant subalpine coniferous tree species at the Niwot Ridge Mountain Research Station forested NEON site in Colorado, USA. We trained pixel-based Random Forest (RF) machine learning models using a series of training data sets along with remote sensing raster data as descriptive features. The highest classification accuracies, 69% and 60% based on internal RF error assessment and an independent validation set, respectively, were obtained using circular tree crown polygons created with half the maximum crown diameter per tree. LiDAR-derived data products were the most important features for species classification, followed by vegetation indices. This work contributes to the open development of well-labeled training data sets for forest composition mapping using openly available NEON data without requiring external data collection, manual delineation steps, or site-specific parameters.

Machine Learning for Gully Feature Extraction Based on a Pan-Sharpened Multispectral Image: Multiclass vs. Binary Approach

Gullies reduce both the quality and quantity of productive land, posing a serious threat to sustainable agriculture, hence, food security. Machine Learning (ML) algorithms are essential tools in the identification of gullies and can assist in strategic decision-making relevant to soil conservation. Nevertheless, accurate identification of gullies is a function of the selected ML algorithms, the image and number of classes used, i.e., binary (two classes) and multiclass. We applied Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Random Forest (RF) on a Systeme Pour l’Observation de la Terre (SPOT-7) image to extract gullies and investigated whether the multiclass (m) approach can offer better classification accuracy than the binary (b) approach. Using repeated k-fold cross-validation, we generated 36 models. Our findings revealed that, of these models, both RFb (98.70%) and SVMm (98.01%) outperformed the LDA in terms of overall accuracy (OA). However, the LDAb (99.51%) recorded the highest producer’s accuracy (PA) but had low corresponding user’s accuracy (UA) with 18.5%. The binary approach was generally better than the multiclass approach; however, on class level, the multiclass approach outperformed the binary approach in gully identification. Despite low spectral resolution, the pan-sharpened SPOT-7 product successfully identified gullies. The proposed methodology is relatively simple, but practically sound, and can be used to monitor gullies within and beyond the study region.

Detecting Dead Standing Eucalypt Trees from Voxelised Full-Waveform Lidar Using Multi-Scale 3D-Windows for Tackling Height and Size Variations

In southern Australia, many native mammals and birds rely on hollows for sheltering, while hollows are more likely to exist on dead trees. Therefore, detection of dead trees could be useful in managing biodiversity. Detecting dead standing (snags) versus dead fallen trees (Coarse Woody Debris—CWD) is a very different task from a classification perspective. This study focuses on improving detection of dead standing eucalypt trees from full-waveform LiDAR. Eucalypt trees have irregular shapes making delineation of them challenging. Additionally, since the study area is a native forest, trees significantly vary in terms of height, density and size. Therefore, we need methods that will be resistant to those challenges. Previous study showed that detection of dead standing trees without tree delineation is possible. This was achieved by using single size 3D-windows to extract structural features from voxelised full-waveform LiDAR and characterise dead (positive samples) and live (negative samples) trees for training a classifier. This paper adds on by proposing the usage of multi-scale 3D-windows for tackling height and size variations of trees. Both the single 3D-windows approach and the new multi-scale 3D-windows approach were implemented for comparison purposes. The accuracy of the results was calculated using the precision and recall parameters and it was proven that the multi-scale 3D-windows approach performs better than the single size 3D-windows approach. This open ups possibilities for applying the proposed approach on other native forest related applications.