Monitoring cross-border trails using airborne digital multispectral imagery and interactive image analysis techniques

The objective of this study was to evaluate image-based procedures for monitoring cross-border foot trails in the US – Mexico border zone in eastern San Diego County using airborne remote sensing techniques. Specifically, digital multi-spectral and multi-temporal imagery from an airborne digital multi-spectral imaging system, digital image processing, and visual image analysis techniques were explored in the context of detecting and delineating new trail features and updating trail GIS layers. Three trail updating approaches: map-to-image (M-I) overlay, map and image-to-image (M/I-I) differencing, map and image-to-image (M/I-I) swiping and two types of spectral transform, PCA and NDVI, were tested and compared. The M-I overlay was found to be the most reliable trail updating approach. The optimal image enhancement method for the M-I overlay approach varied with vegetation structure. PCA imagery yielded better results than NDVI imagery in a highly disturbed area and NDVI imagery performed better in a densely vegetated area. The M/I-I swiping approach was useful for distinguishing misregistered extant trails from new trail features.     

Object-based habitat mapping using very high spatial resolution multispectral and hyperspectral imagery with LiDAR data

This study investigated the combined use of multispectral/hyperspectral imagery and LiDAR data for habitat mapping across parts of south Cumbria, North West England. The methodology adopted in this study integrated spectral information contained in pansharp QuickBird multispectral/AISA Eagle hyperspectral imagery and LiDAR-derived measures with object-based machine learning classifiers and ensemble analysis techniques. Using the LiDAR point cloud data, elevation models (such as the Digital Surface Model and Digital Terrain Model raster) and intensity features were extracted directly. The LiDAR-derived measures exploited in this study included Canopy Height Model, intensity and topographic information (i.e. mean, maximum and standard deviation). These three LiDAR measures were combined with spectral information contained in the pansharp QuickBird and Eagle MNF transformed imagery for image classification experiments. A fusion of pansharp QuickBird multispectral and Eagle MNF hyperspectral imagery with all LiDAR-derived measures generated the best classification accuracies, 89.8 and 92.6% respectively. These results were generated with the Support Vector Machine and Random Forest machine learning algorithms respectively. The ensemble analysis of all three learning machine classifiers for the pansharp QuickBird and Eagle MNF fused data outputs did not significantly increase the overall classification accuracy. Results of the study demonstrate the potential of combining either very high spatial resolution multispectral or hyperspectral imagery with LiDAR data for habitat mapping.     

Most similar neighbor imputation of forest attributes using metrics derived from combined airborne LIDAR and multispectral sensors

In the context of predicting forest attributes using a combination of airborne LIDAR and multispectral (MS) sensors, we suggest the inclusion of normalized difference vegetation index (NDVI) metrics along with the more traditional LIDAR height metrics. Here the data fusion method consists of back-projecting LIDAR returns onto original MS images, avoiding co-registration errors. The prediction method is based on non-parametric imputation (the most similar neighbor). Predictor selection and accuracy assessment include hypothesis tests and over-fitting prevention methods. Results show improvements when using combinations of LIDAR and MS compared to using either of them alone. The MS sensor has little explanatory capacity for forest variables dependent on tree height, already well determined from LIDAR alone. However, there is potential for variables dependent on tree diameters and their density. The combination of LIDAR and MS sensors can be very beneficial for predicting variables describing forests structural heterogeneity, which are best described from synergies between LIDAR heights and NDVI dispersion. Results demonstrate the potential of NDVI metrics to increase prediction accuracy of forest attributes. Their inclusion in the predictor dataset may, however, in a few cases be detrimental to accuracy, and therefore we recommend to carefully assess the possible advantages of data fusion on a case-by-case basis.     

Determination of chlorophyll content of small water bodies (kettle holes) using hyperspectral airborne data

This study presents an approach for chlorophyll content determination of small shallow water bodies (kettle holes) from hyperspectral airborne ROSIS and HyMap data (acquired on 15 May and 29 July 2008 respectively). Investigated field and airborne spectra for almost all kettle holes do not correspond to each other due to differences in ground sampling distance. Field spectra were collected from the height of 30–35 cm (i.e. area of 0.01–0.015 m²). Airborne pixels of ROSIS and HyMap imageries cover an area of 4 m² and 16 m² respectively and their spectra are highly influenced by algae or bottom properties of the kettle holes. Analysis of airborne spectra revealed that chlorophyll absorption near 677 nm is the same for both datasets. In order to enhance absorption properties, both airborne hyperspectral datasets were normalized by the continuum removal approach. Linear regression algorithms for ROSIS and HyMap datasets were derived using normalized average chlorophyll absorption spectra for each kettle hole. Overall accuracy of biomass mapping for ROSIS data was 71%, and for HyMap 64%. Biomass mapping results showed that, depending on the type of kettle hole, algae distribution, the ‘packaging effect’ and bottom reflection lead to miscalculations of the chlorophyll content using hyperspectral airborne data.     

Fusion of airborne LiDAR and multispectral sensors reveals synergic capabilities in forest structure characterization

Forest stand structure is an important concept for ecology and planning in sustainable forest management. In this article, we consider that the incorporation of complementary multispectral information from optical sensors to Light Detection and Ranging (LiDAR) may be advantageous, especially through data fusion by back-projecting the LiDAR points onto the multispectral image. A multivariate data set of both LiDAR and multispectral metrics was related with a multivariate data set of stand structural variables measured in a Scots pine forest through canonical correlation analysis (CCA). Four statistically significant pairs of canonical variables were found, which explained 83.0% accumulated variance. The first pair of canonical variables related indicators of stand development, i.e. height and volume, with LiDAR height metrics. CCA also found attributes describing stand density to be related to LiDAR and spectral variables determining canopy coverage. Other canonical variables pertained to Lorenz curve-derived attributes, which are measures of within-stand tree size variability and heterogeneity, able to discriminate even-sized from uneven-sized stands. The most relevant result was to find that metrics derived from the multispectral sensor showed significant explanatory potential for the prediction of these structural attributes. Therefore, we concluded that metrics derived from the optical sensor have potential for complementing the information from the LiDAR sensor in describing structural properties of forest stands. We recommend the use of back-projecting for jointly exploiting the synergies of both sensors using similar types of metrics as they are customary in forestry applications of LiDAR.     

Extraction, modelling, and use of linear features for restitution of airborne hyperspectral imagery

This paper presents an approach for the restitution of airborne hyperspectral imagery with linear features. The approach consisted of semi-automatic line extraction and mathematical modelling of the linear features. First, the line was approximately determined manually and refined using dynamic programming. The extracted lines could then be used as control data with the ground information of the lines, or as constraints with simple assumption for the ground information of the line. The experimental results are presented numerically in tables of RMS residuals of check points as well as visually in ortho-rectified images.     

机载激光雷达及高光谱的森林乔木物种多样性遥感监测

利用机载LiDAR和高光谱数据并结合37个地面调查样本数据,基于结构差异与光谱变异理论,通过相关分析法分别筛选了3个最优林冠结构参数和6个最优光谱指数,在单木尺度上利用自适应C均值模糊聚类算法,在神农架国家自然保护区开展森林乔木物种多样性监测,实现了森林乔木物种多样性的区域成图。研究结果表明,(1)基于结合形态学冠层控制的分水岭算法可以获得较高精度的单木分割结果(R~2=0.88,RMSE=13.17,P0.001);(2)基于LiDAR数据提取的9个结构参数中,95%百分位高度、冠层盖度和植被穿透率为最优结构参数,与Shannon-Wiener指数的相关性达到R~2=0.39—0.42(P0.01);(3)基于机载高光谱数据筛选的16个常用的植被指数中,CRI、OSAVI、Narrow band NDVI、SR、Vogelmann index1、PRI与Shannon-Wiener指数的相关性最高(R~2=0.37—0.45,P0.01);(4)在研究区,利用以30 m×30 m为窗口的自适应模糊C均值聚类算法可预测的最大森林乔木物种数为20,物种丰富度的预测精度为R~2=0.69,RMSE=3.11,Shannon-Wiener指数的预测精度为R~2=0.70,RMSE=0.32。该研究在亚热带森林开展乔木物种多样性监测,是在区域尺度上进行物种多样性成图的重要实践,可有效补充森林生物多样性本底数据的调查手段,有助于实现生物多样性的长期动态监测及科学分析森林物种多样性的现状和变化趋势。     

结合NSCT和KPCA的高光谱遥感图像去噪

针对高光谱遥感图像易受噪声干扰,本文提出了一种基于非下采样Contourlet变换NSCT(Nonsubsampled Contourlet Transform)和核主成分分析KPCA(Kernel Principal Component Analysis)的去噪方法。首先对高光谱各波段图像进行NSCT分解;然后利用KPCA对NSCT系数进行处理,并在KPCA重构时依据各类噪声的特性选取合适的主成分;最后用处理过的系数进行逆变换得到去噪图像。实验结果表明,本文方法抑制了高光谱遥感图像中的噪声干扰,较完整地保留了原始数据的有效信息。     

Mapping post-fire forest regeneration and vegetation recovery using a combination of very high spatial resolution and hyperspectral satellite imagery

Careful evaluation of forest regeneration and vegetation recovery after a fire event provides vital information useful in land management. The use of remotely sensed data is considered to be especially suitable for monitoring ecosystem dynamics after fire. The aim of this work was to map post-fire forest regeneration and vegetation recovery on the Mediterranean island of Thasos by using a combination of very high spatial (VHS) resolution (QuickBird) and hyperspectral (EO-1 Hyperion) imagery and by employing object-based image analysis. More specifically, the work focused on (1) the separation and mapping of three major post-fire classes (forest regeneration, other vegetation recovery, unburned vegetation) existing within the fire perimeter, and (2) the differentiation and mapping of the two main forest regeneration classes, namely, Pinus brutia regeneration, and Pinus nigra regeneration. The data used in this study consisted of satellite images and field observations of homogeneous regenerated and revegetated areas. The methodology followed two main steps: a three-level image segmentation, and, a classification of the segmented images. The process resulted in the separation of classes related to the aforementioned objectives. The overall accuracy assessment revealed very promising results (approximately 83.7% overall accuracy, with a Kappa Index of Agreement of 0.79). The achieved accuracy was 8% higher when compared to the results reported in a previous work in which only the EO-1 Hyperion image was employed in order to map the same classes. Some classification confusions involving the classes of P. brutia regeneration and P. nigra regeneration were observed. This could be attributed to the absence of large and dense homogeneous areas of regenerated pine trees in the study area.     

局部特征匹配的卫星影像几何偏差估计可行性分析

几何定位偏差是评估卫星影像质量的重要参数之一。本文利用特征匹配方法进行卫星影像几何偏差估计,并对该方法的可行性进行分析。首先,选取包含典型地物的局部卫星影像构建路标基准影像库;其次,采用局部点特征描述子对卫星影像与路标基准影像库进行亚像素提取与匹配,确定同名像点,然后为提升同名像点匹配精度,提出了顾及空间关系与几何一致性约束的匹配策略,继而,对卫星影像与路标基准影像库的匹配像点进行内插计算,获取同名像点的地理空间坐标;最后,通过差值计算得到卫星影像相对路标基准影像的偏移参数,完成几何偏差估计和可行性分析。本文以我国风云卫星为例,选取包含海岸线、湖泊、山脉、河流及岛屿的局部卫星影像集构建路标基准影像库,并选用SIFT、SURF、ORB 3个局部特征进行测试分析。测试结果表明,基于局部特征匹配策略对卫星影像几何偏差估计的系统误差小于0.1像素,定量地证明了局部特征匹配方法对卫星影像几何偏差估计的可行性。     

3D geostatistical modelling and uncertainty analysis of clay minerals distribution in reservoir rocks

Drilling through shale layers is associated with many problems such as wellbore instability, hole wash out, pipe stuck, etc. Reactive shale with swelling potential can lead to such problems. Reactive shale is a shale that has high swelling clay mineral like montmorillonite although there was a little amount of this clay mineral. So determine the type of clay mineral in shale formation and obtain the distribution of swellable clay mineral can help us in the selection of stable well path and determination of well location that has least instability problem, prevention and prediction of problem associated with swelling clay mineral such as pipe stuck, hole wash out, bit balling, etc. In this paper, we used thorium and potassium from corrected gamma ray (CGR) log to obtain the type and percentage of clay mineral. Then used X-ray diffraction for more detail analysis of clay mineral constitute. Finally, geostatistical modelling is used to obtain the continuity of clay mineral in the reservoir and estimate the amount of desire clay mineral along the reservoir.     

Relevance of airborne lidar and multispectral image data for urban scene classification using Random Forests

Airborne lidar systems have become a source for the acquisition of elevation data. They provide georeferenced, irregularly distributed 3D point clouds of high altimetric accuracy. Moreover, these systems can provide for a single laser pulse, multiple returns or echoes, which correspond to different illuminated objects. In addition to multi-echo laser scanners, full-waveform systems are able to record 1D signals representing a train of echoes caused by reflections at different targets. These systems provide more information about the structure and the physical characteristics of the targets. Many approaches have been developed, for urban mapping, based on aerial lidar solely or combined with multispectral image data. However, they have not assessed the importance of input features. In this paper, we focus on a multi-source framework using aerial lidar (multi-echo and full waveform) and aerial multispectral image data. We aim to study the feature relevance for dense urban scenes. The Random Forests algorithm is chosen as a classifier: it runs efficiently on large datasets, and provides measures of feature importance for each class. The margin theory is used as a confidence measure of the classifier, and to confirm the relevance of input features for urban classification. The quantitative results confirm the importance of the joint use of optical multispectral and lidar data. Moreover, the relevance of full-waveform lidar features is demonstrated for building and vegetation area discrimination.     

Comparative analysis of different retrieval methods for mapping grassland leaf area index using airborne imaging spectroscopy

Fine scale maps of vegetation biophysical variables are useful status indicators for monitoring and managing national parks and endangered habitats. Here, we assess in a comparative way four different retrieval methods for estimating leaf area index (LAI) in grassland: two radiative transfer model (RTM) inversion methods (one based on look-up-tables (LUT) and one based on predictive equations) and two statistical modelling methods (one partly, the other entirely based on in situ data). For prediction, spectral data were used that had been acquired over Majella National Park in Italy by the airborne hyperspectral HyMap instrument. To assess the performance of the four investigated models, the normalized root mean squared error (nRMSE) and coefficient of determination (R²) between estimates and in situ LAI measurements are reported (n = 41). Using a jackknife approach, we also quantified the accuracy and robustness of empirical models as a function of the size of the available calibration data set. The results of the study demonstrate that the LUT-based RTM inversion yields higher accuracies for LAI estimation (R² = 0.91, nRMSE = 0.18) as compared to RTM inversions based on predictive equations (R² = 0.79, nRMSE = 0.38). The two statistical methods yield accuracies similar to the LUT method. However, as expected, the accuracy and robustness of the statistical models decrease when the size of the calibration database is reduced to fewer samples. The results of this study are of interest for the remote sensing community developing improved inversion schemes for spaceborne hyperspectral sensors applicable to different vegetation types. The examples provided in this paper may also serve as illustrations for the drawbacks and advantages of physical and empirical models.     

Individual tree crown approach for predicting site index in boreal forests using airborne laser scanning and hyperspectral data

Site productivity is essential information for sustainable forest management and site index (SI) is the most common quantitative measure of it. The SI is usually determined for individual tree species based on tree height and the age of the 100 largest trees per hectare according to stem diameter. The present study aimed to demonstrate and validate a methodology for the determination of SI using remotely sensed data, in particular fused airborne laser scanning (ALS) and airborne hyperspectral data in a forest site in Norway. The applied approach was based on individual tree crown (ITC) delineation: tree species, tree height, diameter at breast height (DBH), and age were modelled and predicted at ITC level using 10-fold cross validation. Four dominant ITCs per 400 m² plot were selected as input to predict SI at plot level for Norway spruce (Picea abies (L.) Karst.) and Scots pine (Pinus sylvestris L.). We applied an experimental setup with different subsets of dominant ITCs with different combinations of attributes (predicted or field-derived) for SI predictions. The results revealed that the selection of the dominant ITCs based on the largest DBH independent of tree species, predicted the SI with similar accuracy as ITCs matched with field-derived dominant trees (RMSE: 27.6% vs 23.3%). The SI accuracies were at the same level when dominant species were determined from the remotely sensed or field data (RMSE: 27.6% vs 27.8%). However, when the predicted tree age was used the SI accuracy decreased compared to field-derived age (RMSE: 27.6% vs 7.6%). In general, SI was overpredicted for both tree species in the mature forest, while there was an underprediction in the young forest. In conclusion, the proposed approach for SI determination based on ITC delineation and a combination of ALS and hyperspectral data is an efficient and stable procedure, which has the potential to predict SI in forest areas at various spatial scales and additionally to improve existing SI maps in Norway.     

Estimating chlorophyll with thermal and broadband multispectral high resolution imagery from an unmanned aerial system using relevance vector machines for precision agriculture

Precision agriculture requires high-resolution information to enable greater precision in the management of inputs to production. Actionable information about crop and field status must be acquired at high spatial resolution and at a temporal frequency appropriate for timely responses. In this study, high spatial resolution imagery was obtained through the use of a small, unmanned aerial system called AggieAir^TM. Simultaneously with the AggieAir flights, intensive ground sampling for plant chlorophyll was conducted at precisely determined locations. This study reports the application of a relevance vector machine coupled with cross validation and backward elimination to a dataset composed of reflectance from high-resolution multi-spectral imagery (VIS–NIR), thermal infrared imagery, and vegetative indices, in conjunction with in situ SPAD measurements from which chlorophyll concentrations were derived, to estimate chlorophyll concentration from remotely sensed data at 15-cm resolution. The results indicate that a relevance vector machine with a thin plate spline kernel type and kernel width of 5.4, having LAI, NDVI, thermal and red bands as the selected set of inputs, can be used to spatially estimate chlorophyll concentration with a root-mean-squared-error of 5.31 μg cm⁻², efficiency of 0.76, and 9 relevance vectors.     

Estimation of green grass/herb biomass from airborne hyperspectral imagery using spectral indices and partial least squares regression

The main objective was to determine whether partial least squares (PLS) regression improves grass/herb biomass estimation when compared with hyperspectral indices, that is normalised difference vegetation index (NDVI) and red-edge position (REP). To achieve this objective, fresh green grass/herb biomass and airborne images (HyMap) were collected in the Majella National Park, Italy in the summer of 2005. The predictive performances of hyperspectral indices and PLS regression models were then determined and compared using calibration (n = 30) and test (n = 12) data sets. The regression model derived from NDVI computed from bands at 740 and 771 nm produced a lower standard error of prediction (SEP = 264 g m⁻²) on the test data compared with the standard NDVI involving bands at 665 and 801 nm (SEP = 331 g m⁻²), but comparable results with REPs determined by various methods (SEP = 261 to 295 g m⁻²). PLS regression models based on original, derivative and continuum-removed spectra produced lower prediction errors (SEP = 149 to 256 g m⁻²) compared with NDVI and REP models. The lowest prediction error (SEP = 149 g m⁻², 19% of mean) was obtained with PLS regression involving continuum-removed bands. In conclusion, PLS regression based on airborne hyperspectral imagery provides a better alternative to univariate regression involving hyperspectral indices for grass/herb biomass estimation in the Majella National Park.     

Detecting Sirex noctilio grey-attacked and lightning-struck pine trees using airborne hyperspectral data,random forest and support vector machines classifiers

The visual progression of sirex (Sirex noctilio) infestation symptoms has been categorized into three distinct infestation phases, namely the green, red and grey stages. The grey stage is the final stage which leads to almost complete defoliation resulting in dead standing trees or snags. Dead standing pine trees however, could also be due to the lightning damage. Hence, the objective of the present study was to distinguish amongst healthy, sirex grey-attacked and lightning-damaged pine trees using AISA Eagle hyperspectral data, random forest (RF) and support vector machines (SVM) classifiers. Our study also presents an opportunity to look at the possibility of separating amongst the previously mentioned pine trees damage classes and other landscape classes on the study area. The results of the present study revealed the robustness of the two machine learning classifiers with an overall accuracy of 74.50% (total disagreement = 26%) for RF and 73.50% (total disagreement = 27%) for SVM using all the remaining AISA Eagle spectral bands after removing the noisy ones. When the most useful spectral bands as measured by RF were exploited, the overall accuracy was considerably improved; 78% (total disagreement = 22%) for RF and 76.50% (total disagreement = 24%) for SVM. There was no significant difference between the performances of the two classifiers as demonstrated by the results of McNemar’s test (chi-squared; χ² = 0.14, and 0.03 when all the remaining ASIA Eagle wavebands, after removing the noisy ones and the most important wavebands were used, respectively). This study concludes that AISA Eagle data classified using RF and SVM algorithms provide relatively accurate information that is important to the forest industry for making informed decision regarding pine plantations health protocols.     

利用Landsat卫星影像进行东莞市地表温度反演与变化分析

随着快速城市化进程,我国东南沿海一些城市新增了大量的城市建设用地,影响到了城市的人居环境,诱发了城市热的环境问题.探索城市热环境时空布局与变化特征,分析变化原因对城市可持续发展和人居环境质量改善具有重要的现实意义.本文以东莞市为例,以1986—2017年间8个时相Landsat系列卫星影像为数据源,利用单通道算法对东莞...     

Modeling biophysical properties of broad-leaved stands in the hyrcanian forests of Iran using fused airborne laser scanner data and ultraCam-D images

Inventories of mixed broad-leaved forests of Iran mainly rely on terrestrial measurements. Due to rapid changes and disturbances and great complexity of the silvicultural systems of these multilayer forests, frequent repetition of conventional ground-based plot surveys is often cost prohibitive. Airborne laser scanning (ALS) and multispectral data offer an alternative or supplement to conventional inventories in the Hyrcanian forests of Iran. In this study, the capability of a combination of ALS and UltraCam-D data to model stand volume, tree density, and basal area using random forest (RF) algorithm was evaluated. Systematic sampling was applied to collect field plot data on a 150 m × 200 m sampling grid within a 1100 ha study area located at 36°38′- 36°42′N and 54°24′–54°25′E. A total of 308 circular plots (0.1 ha) were measured for calculation of stand volume, tree density, and basal area per hectare. For each plot, a set of variables was extracted from both ALS and multispectral data. The RF algorithm was used for modeling of the biophysical properties using ALS and UltraCam-D data separately and combined. The results showed that combining the ALS data and UltraCam-D images provided a slight increase in prediction accuracy compared to separate modeling. The RMSE as percentage of the mean, the mean difference between observed and predicted values, and standard deviation of the differences using a combination of ALS data and UltraCam-D images in an independent validation at 0.1-ha plot level were 31.7%, 1.1%, and 84 m³ ha⁻¹ for stand volume; 27.2%, 0.86%, and 6.5 m² ha⁻¹ for basal area, and 35.8%, −4.6%, and 77.9 n ha⁻¹ for tree density, respectively. Based on the results, we conclude that fusion of ALS and UltraCam-D data may be useful for modeling of stand volume, basal area, and tree density and thus gain insights into structural characteristics in the complex Hyrcanian forests.     

Comparative performance of convolutional neural network,weighted and conventional support vector machine and random forest for classifying tree species using hyperspectral and photogrammetric data

ABSTRACT

The classification of tree species can significantly benefit from high spatial and spectral information acquired by unmanned aerial vehicles (UAVs) associated with advanced classification methods. This study investigated the following topics concerning the classification of 16 tree species in two subtropical forest fragments of Southern Brazil: i) the potential integration of UAV-borne hyperspectral images with 3D information derived from their photogrammetric point cloud (PPC); ii) the performance of two machine learning methods (support vector machine – SVM and random forest – RF) when employing different datasets at a pixel and individual tree crown (ITC) levels; iii) the potential of two methods for dealing with the imbalanced sample set problem: a new weighted SVM (wSVM) approach, which attributes different weights to each sample and class, and a deep learning classifier (convolutional neural network – CNN), associated with a previous step to balance the sample set; and finally, iv) the potential of this last classifier for tree species classification as compared to the above mentioned machine learning methods. Results showed that the inclusion of the PPC features to the hyperspectral data provided a great accuracy increase in tree species classification results when conventional machine learning methods were applied, between 13 and 17% depending on the classifier and the study area characteristics. When using the PPC features and the canopy height model (CHM), associated with the majority vote (MV) rule, the SVM, wSVM and RF classifiers reached accuracies similar to the CNN, which outperformed these classifiers for both areas when considering the pixel-based classifications (overall accuracy of 84.4% in Area 1, and 74.95% in Area 2). The CNN was between 22% and 26% more accurate than the SVM and RF when only the hyperspectral bands were employed. The wSVM provided a slight increase in accuracy not only for some lesser represented classes, but also some major classes in Area 2. While conventional machine learning methods are faster, they demonstrated to be less stable to changes in datasets, depending on prior segmentation and hand-engineered features to reach similar accuracies to those attained by the CNN. To date, CNNs have been barely explored for the classification of tree species, and CNN-based classifications in the literature have not dealt with hyperspectral data specifically focusing on tropical environments. This paper thus presents innovative strategies for classifying tree species in subtropical forest areas at a refined legend level, integrating UAV-borne 2D hyperspectral and 3D photogrammetric data and relying on both deep and conventional machine learning approaches.