Search Results (7)

This study used Jilin-1 satellite data and unmanned aerial vehicle (UAV)-collected visible-thermal infrared imagery to calculate twelve spectral indices and evaluate their effectiveness in distinguishing post-fire forest areas and identifying human-altered land-cover changes in Jinyun Mountain, Chongqing. The research goals included mapping wildfire impacts with M-statistic separability, measuring land-cover distinguishability through Jeffries–Matusita (JM) distance analysis, classifying land-cover types using the random forest (RF) algorithm, and verifying classification accuracy. Cumulative human disturbances—such as land clearing, replanting, and road construction—significantly blocked the natural recovery of burn scars, and during long-term human-assisted recovery periods over one year, the Red Green Blue Index (RGBI), Green Leaf Index (GLI), and Excess Green Index (EXG) showed high classification accuracy for six land-cover types: road, bare soil, deadwood, bamboo, broadleaf, and grass. Key accuracy measures showed producer accuracy (PA) > 0.8, user accuracy (UA) > 0.8, overall accuracy (OA) > 90%, and a kappa coefficient > 0.85. Validation results confirmed that visible-spectrum indices are good at distinguishing photosynthetic vegetation, thermal bands help identify artificial surfaces, and combined thermal-visible indices solve spectral confusion in deadwood recognition. Spectral indices provide high-precision quantitative evidence for monitoring post-fire land-cover changes, especially under human intervention, thus offering important data support for time-based modeling of post-fire forest recovery and improvement of ecological restoration plans. Full article

We present an evaluation of different deep learning and machine learning approaches for tree health classification in the Black Forest, the Harz Mountains, and the Göttinger Forest on a unique, highly accurate tree-level dataset. The multispectral UAV data were collected from eight forest plots with diverse tree species, mostly conifers. As ground truth data (GTD), nearly 1500 tree polygons with related attribute information on the health status of the trees were used. This data were collected during extensive fieldwork using a mobile application and subsequent individual tree segmentation. Extensive preprocessing included normalization, NDVI calculations, data augmentation to deal with the underrepresented classes, and splitting the data into training, validation, and test sets. We conducted several experiments using a classical machine learning approach (random forests), as well as different convolutional neural networks (CNNs)—ResNet50, ResNet101, VGG16, and Inception-v3—on different datasets and classes to evaluate the potential of these algorithms for tree health classification. Our first experiment was a binary classifier of healthy and damaged trees, which did not consider the degree of damage or tree species. The best results of a 0.99 test accuracy and an F1 score of 0.99 were obtained with ResNet50 on four band composites using the red, green, blue, and infrared bands (RGBI images), while VGG16 had the worst performance, with an F1 score of only 0.78. In a second experiment, we also distinguished between coniferous and deciduous trees. The F1 scores ranged from 0.62 to 0.99, with the highest results obtained using ResNet101 on derived vegetation indices using the red edge band of the camera (NDVIre images). Finally, in a third experiment, we aimed at evaluating the degree of damage: healthy, slightly damaged, and medium or heavily damaged trees. Again, ResNet101 had the best performance, this time on RGBI images with a test accuracy of 0.98 and an average F1 score of 0.97. These results highlight the potential of CNNs to handle high-resolution multispectral UAV data for the early detection of damaged trees when good training data are available. Full article

Disease and pest detection of grape foliage is essential for grape yield and quality. RGB image (RGBI), multispectral image (MSI), and thermal infrared image (TIRI) are widely used in the health detection of plants. In this study, we collected three types of grape foliage images with six common classes (anthracnose, downy mildew, leafhopper, mites, viral disease, and healthy) in the field. ShuffleNet V2 was used to build up detection models. According to the accuracy of RGBI, MSI, TIRI, and multi-source data concatenation (MDC) models, and a multi-source data fusion (MDF) decision-making method was proposed for improving the detection performance for grape foliage, aiming to enhance the decision-making for RGBI of grape foliage by fusing the MSI and TIRI. The results showed that 40% of the incorrect detection outputs were rectified using the MDF decision-making method. The overall accuracy of MDF model was 96.05%, which had improvements of 2.64%, 13.65%, and 27.79%, compared with the RGBI, MSI, and TIRI models using label smoothing, respectively. In addition, the MDF model was based on the lightweight network with 3.785 M total parameters and 0.362 G multiply-accumulate operations, which could be highly portable and easy to be applied. Full article

Forest species classifications are becoming increasingly automated as advances are made in machine learning. Complex algorithms can reach high accuracies, but are not always suitable for small-scale classifications, which may benefit from simpler conventional methods. The goal of this classification was to identify contiguous stands of ponderosa pine (Pinus ponderosa Douglas ex Lawson) against a mix of forest and non-forest background in the southern Willamette Valley, Oregon. The study area is approximately 816,600 ha, considerably larger than most study areas used for presenting techniques for tree species classification. To achieve the objective, we used two classification procedures, one parametric and one non-parametric. For the parametric method, we selected the maximum likelihood (ML) algorithm, whereas for the non-parametric method we chose the random forest (RF) algorithm. To identify ponderosa pine, we used 1 m spatial resolution red-green-blue-infrared (RGBI) aerial images supplied by the U.S. National Agriculture Imagery Program (NAIP) and 1 m spatial resolution canopy height models (CHMs) provided by the Oregon Department of Geology and Mineral Industries (DOGAMI). We tested four data variations for each method: Aerial imagery, CHM-masked aerial imagery, aerial imagery with an additional CHM band, and CHM-masked aerial imagery with a CHM band. The parametric classifications of aerial imagery alone reached an average kappa coefficient of 0.29, which increased to 0.51 when masked with CHM data. The incorporation of CHM data as a fifth band resulted in a similar improvement in kappa (0.47), but the most effective parametric method was the incorporation of CHM data as both a fifth band and a post-classification mask, resulting in a kappa coefficient of 0.89. The non-parametric classification of aerial imagery achieved a mean validation kappa coefficient of 0.85 collectively and 0.90 individually, which only increased by approximately 0.01 or less when the CHM masks were applied. The addition of the CHM band increased the kappa value to 0.91 for both individual and collective tile classifications. The highest kappa of all methods was achieved through five-band non-parametric classification with the addition of the CHM band (0.94) for both collective and individual classifications. Our results suggest that parametric methods, when enhanced with a CHM mask, could be suitable for large-area, small-scale classifications based on RGBI imagery, but a non-parametric classification of fused spectral and height data will generally achieve the highest accuracy for large, unbalanced datasets. Full article

Deadwood mapping is of high relevance for studies on forest biodiversity, forest disturbance, and dynamics. As deadwood predominantly occurs in forests characterized by a high structural complexity and rugged terrain, the use of remote sensing offers numerous advantages over terrestrial inventory. However, deadwood misclassifications can occur in the presence of bare ground, displaying a similar spectral signature. In this study, we tested the potential to detect standing deadwood (h > 5 m) using orthophotos (0.5 m resolution) and digital surface models (DSM) (1 m resolution), both derived from stereo aerial image matching (0.2 m resolution and 60%/30% overlap (end/side lap)). Models were calibrated in a 600 ha mountain forest area that was rich in deadwood in various stages of decay. We employed random forest (RF) classification, followed by two approaches for addressing the deadwood-bare ground misclassification issue: (1) post-processing, with a mean neighborhood filter for “deadwood”-pixels and filtering out isolated pixels and (2) a “deadwood-uncertainty” filter, quantifying the probability of a “deadwood”-pixel to be correctly classified as a function of the environmental and spectral conditions in its neighborhood. RF model validation based on data partitioning delivered high user’s (UA) and producer’s (PA) accuracies (both > 0.9). Independent validation, however, revealed a high commission error for deadwood, mainly in areas with bare ground (UA = 0.60, PA = 0.87). Post-processing (1) and the application of the uncertainty filter (2) improved the distinction between deadwood and bare ground and led to a more balanced relation between UA and PA (UA of 0.69 and 0.74, PA of 0.79 and 0.80, under (1) and (2), respectively). Deadwood-pixels showed 90% location agreement with manually delineated reference to deadwood objects. With both alternative solutions, deadwood mapping achieved reliable results and the highest accuracies were obtained with deadwood-uncertainty filter. Since the information on surface heights was crucial for correct classification, enhancing DSM quality could substantially improve the results. Full article

More than 50% of the national lands in Japan have been surveyed by airborne laser scanning (ALS) data with different point densities; and developing an effective approach to take full advantage of these ALS data for forest management has thus become an urgent topic of study. This study attempted to assess the utility of ALS data for individual tree detection and species classification in a mixed forest with a high canopy density. For comparison, two types of tree tops and tree crowns in the study area were delineated by the individual tree crown (ITC) approach using the green band of the orthophoto imagery and the digital canopy height model (DCHM) derived from the ALS data, respectively. Then, the two types of tree crowns were classified into four classes—Pinus densiflora (Pd), Chamaecyparis obtusa (Co), Larix kaempferi (Lk), and broadleaved trees (Bl)—by a crown-based classification approach using different combinations of the three orthophoto bands with intensity and slope maps as follows: RGB (red, green and blue); RGB and intensity (RGBI); RGB and slope (RGBS); and RGB, intensity and slope (RGBIS). Finally, the tree tops were annotated with species attributes from the two best-classified tree crown maps, and the number of different tree species in each compartment was counted for comparison with the field data. The results of our study suggest that the combination of RGBIS yielded greater classification accuracy than the other combinations. In the tree crown classifications delineated by the green band and DCHM data, the improvements in the overall accuracy compared to the RGB ranged from 5.7% for the RGBS to 9.0% for the RGBIS and from 8.3% for the RGBS to 11.8% for the RGBIS. The laser intensity and slope derived from the ALS data may be valuable sources of information for tree species classification, and in terms of distinguishing species for the detection of individual trees, the findings of this study demonstrate the advantages of using DCHM instead of optical data to delineate tree crowns. In conclusion, the synthesis of individual tree delineation using DCHM data and species classification using the RGBIS combination is recommended for interpreting forest resources in the study area. However, the usefulness of this approach must be verified in future studies through its application to other forests. Full article

Forest gaps are important structural elements in forest ecology to which various conservation-relevant, photophilic species are associated. To automatically map forest gaps and detect their changes over time, we developed a method based on Digital Surface Models (DSM) derived from stereoscopic aerial imagery and a LiDAR-based Digital Elevation Model (LiDAR DEM). Gaps were detected and delineated in relation to height and cover of the surrounding forest comparing data from two public flight campaigns (2009 and 2012) in a 1023-ha model region in the Northern Black Forest, Southwest Germany. The method was evaluated using an independent validation dataset obtained by visual stereo-interpretation. Gaps were automatically detected with an overall accuracy of 0.90 (2009) and 0.82 (2012). However, a very high users’ accuracy of more than 0.95 (both years) was counterbalanced by a producer’s accuracy of 0.84 (2009) and 0.73 (2012) as some gaps were not automatically detected. Accuracy was mainly dependent on the shadow occurrence and height of the surrounding forest with user’s accuracies dropping to 0.70 (2009) and 0.52 (2012) in high stands (>8 m tree height). As one important step in the workflow, the class of open forest, an important feature for many forest species, was delineated with a very good overall accuracy of 0.92 (both years) with uncertainties occurring mostly in areas with intermediate canopy cover. Presence of complete or partial shadow and geometric limitations of stereo image matching were identified as the main sources of errors in the method performance, suggesting that images with a higher overlap and resolution and ameliorated image-matching algorithms provide the greatest potential for improvement. Full article