Unmanned Aerial Vehicle Imagery Research Articles

Assessment of Using UAV Imagery over Featureless Surfaces for Topographic Applications

The growth and development of Unmanned Aerial Vehicles (UAVs) as a photogrammetric platform, concurrently with the advances in Computer Vision (CV) and image processing algorithms have resulted using UAV Photogrammetry in several topographic applications. CV software algorithms rely on extracting, describing, and matching tie points from the sequences overlapping images to generate 3D colored point clouds. One of the biggest problems obstructing the automated processing of UAV imagery is the featureless of the covered surface. This paper has provided the ability, results, and accuracy of processing images captured by UAVs over non-textured sandy surfaces by providing four aligning and geo-referencing techniques. These four methods, IG/blind matching, IG/reference matching, DG/blind matching, and DG/reference matching, have been presented and tested for 630 aerial images with 80 % overlap and 80 % side lap covered approximately 1 km2 at altitude 178 m above ground level (AGL). The results showed that the captured images could be used to extract the photogrammetric topographical measurements with reliable accuracy. The four techniques' geometric accuracy has ranged between (0.043 m to 0.076 m) & (0.047 m to 0.074) for generated point clouds and linear exterior orientation (EO) parameters, respectively. The indirect geo-referencing with reference matching (IG/reference) recorded the highest-level accuracy of point clouds with 0.043m RMSE compared to the direct geo-referencing with reference matching (DG/reference) which gave the highest geometric accuracy of the linear EO parameters with 0.047m

Semi-Supervised Learning for Forest Fire Segmentation Using UAV Imagery

Unmanned aerial vehicles (UAVs) are an efficient tool for monitoring forest fire due to its advantages, e.g., cost-saving, lightweight, flexible, etc. Semantic segmentation can provide a model aircraft to rapidly and accurately determine the location of a forest fire. However, training a semantic segmentation model requires a large number of labeled images, which is labor-intensive and time-consuming to generate. To address the lack of labeled images, we propose, in this paper, a semi-supervised learning-based segmentation network, SemiFSNet. By taking into account the unique characteristics of UAV-acquired imagery of forest fire, the proposed method first uses occlusion-aware data augmentation for labeled data to increase the robustness of the trained model. In SemiFSNet, a dynamic encoder network replaces the ordinary convolution with dynamic convolution, thus enabling the learned feature to better represent the fire feature with varying size and shape. To mitigate the impact of complex scene background, we also propose a feature refinement module by integrating an attention mechanism to highlight the salient feature information, thus improving the performance of the segmentation network. Additionally, consistency regularization is introduced to exploit the rich information that unlabeled data contain, thus aiding the semi-supervised learning. To validate the effectiveness of the proposed method, extensive experiments were conducted on the Flame dataset and Corsican dataset. The experimental results show that the proposed model outperforms state-of-the-art methods and is competitive to its fully supervised learning counterpart.

A Mixed Methods Approach for Fuel Characterisation in Gorse (Ulex europaeus L.) Scrub from High-Density UAV Laser Scanning Point Clouds and Semantic Segmentation of UAV Imagery

The classification and quantification of fuel is traditionally a labour-intensive, costly and often subjective operation, especially in hazardous vegetation types, such as gorse (Ulex europaeus L.) scrub. In this study, unmanned aerial vehicle (UAV) technologies were assessed as an alternative to traditional field methodologies for fuel characterisation. UAV laser scanning (ULS) point clouds were captured, and a variety of spatial and intensity metrics were extracted from these data. These data were used as predictor variables in models describing destructively and non-destructively sampled field measurements of total above ground biomass (TAGB) and above ground available fuel (AGAF). Multiple regression of the structural predictor variables yielded correlations of R2 = 0.89 and 0.87 for destructively sampled measurements of TAGB and AGAF, respectively, with relative root mean square error (RMSE) values of 18.6% and 11.3%, respectively. The best metrics for non-destructive field-measurements yielded correlations of R2 = 0.50 and 0.49, with RMSE values of 40% and 30.8%, for predicting TAGB and AGAF, respectively, indicating that ULS-derived structural metrics offer higher levels of precision. UAV-derived versions of the field metrics (overstory height and cover) predicted TAGB and AGAF with R2 = 0.44 and 0.41, respectively, and RMSE values of 34.5% and 21.7%, demonstrating that even simple metrics from a UAV can still generate moderate correlations. In further analyses, UAV photogrammetric data were captured and automatically processed using deep learning in order to classify vegetation into different fuel categories. The results yielded overall high levels of precision, recall and F1 score (0.83 for each), with minimum and maximum levels per class of F1 = 0.70 and 0.91. In conclusion, these ULS-derived metrics can be used to precisely estimate fuel type components and fuel load at fine spatial resolutions over moderate-sized areas, which will be useful for research, wildfire risk assessment and fuel management operations.

Experimental Solution for Estimating Pedestrian Locations from UAV Imagery

This research describes an experimental solution used for estimating the positions of pedestrians from video recordings. Additionally, clustering algorithms were utilized to interpret the data. The system employs the You Only Look Once (YOLO) algorithm for object detection. The detection algorithm is applied to video recordings provided by an unmanned aerial vehicle (UAV). An experimental method for calculating the pedestrian’s geolocation is proposed. The output of the calculation, i.e., the data file, can be visualized on a map and analyzed using cluster analyses, including K-means, DBSCAN, and OPTICS algorithms. The experimental software solution can be deployed on a UAV or other computing devices. Further testing was performed to evaluate the suitability of the selected algorithms and to identify optimal use cases. This solution can successfully detect groups of pedestrians from video recordings and it provides tools for subsequent cluster analyses.

Deep learning-based early weed segmentation using motion blurred UAV images of sorghum fields

Weeds are undesired plants in agricultural fields that affect crop yield and quality by competing for nutrients, water, sunlight and space. For centuries, farmers have used several strategies and resources to remove weeds. The use of herbicide is still the most common control strategy. To reduce the amount of herbicide and impact caused by uniform spraying, site-specific weed management (SSWM) through variable rate herbicide application and mechanical weed control have long been recommended. To implement such precise strategies, accurate detection and classification of weeds in crop fields is a crucial first step. Due to the phenotypic similarity between some weeds and crops as well as changing weather conditions, it is challenging to design an automated system for general weed detection. For efficiency, unmanned aerial vehicles (UAV) are commonly used for image capturing. However, high wind pressure and different drone settings have a severe effect on the capturing quality, what potentially results in degraded images, e.g., due to motion blur. In this paper, we investigate the generalization capabilities of Deep Learning methods for early weed detection in sorghum fields under such challenging capturing conditions. For this purpose, we developed weed segmentation models using three different state-of-the-art Deep Learning architectures in combination with residual neural networks as feature extractors.We further publish a manually annotated and expert-curated UAV imagery dataset for weed detection in sorghum fields under challenging conditions. Our results show that our trained models generalize well regarding the detection of weeds, even for degraded captures due to motion blur. An UNet-like architecture with a ResNet-34 feature extractor achieved an F1-score of over 89% on a hold-out test-set. Further analysis indicate that the trained model performed well in predicting the general plant shape, while most misclassifications appeared at borders of the plants. Beyond that, our approach can detect intra-row weeds without additional information as well as partly occluded plants in contrast to existing research.All data, including the newly generated and annotated UAV imagery dataset, and code is publicly available on GitHub: https://github.com/grimmlab/UAVWeedSegmentation and Mendeley Data: https://doi.org/10.17632/4hh45vkp38.4.

Prediction of Strawberry Dry Biomass from UAV Multispectral Imagery Using Multiple Machine Learning Methods

Biomass is a key biophysical parameter for precision agriculture and plant breeding. Fast, accurate and non-destructive monitoring of biomass enables various applications related to crop growth. In this paper, strawberry dry biomass weight was modeled using 4 canopy geometric parameters (area, average height, volume, standard deviation of height) and 25 spectral variables (5 band original reflectance values and 20 vegetation indices (VIs)) extracted from the Unmanned Aerial Vehicle (UAV) multispectral imagery. Six regression techniques—multiple linear regression (MLR), random forest (RF), support vector machine (SVM), multivariate adaptive regression splines (MARS), eXtreme Gradient Boosting (XGBoost) and artificial neural network (ANN)—were employed and evaluated for biomass prediction. The ANN had the highest accuracy in a five-fold cross-validation, with R2 of 0.89~0.93, RMSE of 7.16~8.98 g and MAE of 5.06~6.29 g. As for the other five models, the addition of VIs increased the R2 from 0.77~0.80 to 0.83~0.86, and reduced the RMSE from 8.89~9.58 to 7.35~8.09 g and the MAE from 6.30~6.70 to 5.25~5.47 g, respectively. Red-edge-related VIs, including the normalized difference red-edge index (NDRE), simple ratio vegetation index red-edge (SRRedEdge), modified simple ratio red-edge (MSRRedEdge) and chlorophyll index red and red-edge (CIred&RE), were the most influential VIs for biomass modeling. In conclusion, the combination of canopy geometric parameters and VIs obtained from the UAV imagery was effective for strawberry dry biomass estimation using machine learning models.

Machine Learning-Based Processing of Multispectral and RGB UAV Imagery for the Multitemporal Monitoring of Vineyard Water Status

The development of unmanned aerial vehicles (UAVs) and light sensors has required new approaches for high-resolution remote sensing applications. High spatial and temporal resolution spectral data acquired by multispectral and conventional cameras (or red, green, blue (RGB) sensors) onboard UAVs can be useful for plant water status determination and, as a consequence, for irrigation management. A study in a vineyard located in south-eastern Spain was carried out during the 2018, 2019, and 2020 seasons to assess the potential uses of these techniques. Different water qualities and irrigation application start throughout the growth cycle were imposed. Flights with RGB and multispectral cameras mounted on a UAV were performed throughout the growth cycle, and orthoimages were generated. These orthoimages were segmented to include only vegetation and calculate the green canopy cover (GCC). The stem water potential was measured, and the water stress integral (Sψ) was obtained during each irrigation season. Multiple linear regression techniques and artificial neural networks (ANNs) models with multispectral and RGB bands, as well as GCC, as inputs, were trained and tested to simulate the Sψ. The results showed that the information in the visible domain was highly related to the Sψ in the 2018 season. For all the other years and combinations of years, multispectral ANNs performed slightly better. Differences in the spatial resolution and radiometric quality of the RGB and multispectral geomatic products explain the good model performances with each type of data. Additionally, RGB cameras cost less and are easier to use than multispectral cameras, and RGB images are simpler to process than multispectral images. Therefore, RGB sensors are a good option for use in predicting entire vineyard water status. In any case, field punctual measurements are still required to generate a general model to estimate the water status in any season and vineyard.

UAVformer: A Composite Transformer Network for Urban Scene Segmentation of UAV Images

Urban scenes segmentation based on UAV (Unmanned aerial vehicle) view is a fundamental task for the applications of smart city such as city planning, land use monitoring, traffic monitoring, and crowd estimation. While urban scenes in UAV image characteristic by large scale variation of objects size and complexity background, which posed challenges to urban scenes segmentation of UAV image. The feature extracting backbone of existing networks cannot extract complex features of UAV image effectively, which limits the performance of urban scenes segmentation. To design segmentation network capable of extracting features of large scale variation urban ground scenes, this study proposed a novel composite transformer network for urban scenes segmentation of UAV image. A composite backbone with aggregation windows multi-head self-attention transformer blocks is proposed to make the extracted features more representatives by adaptive multi-level features fusion, and the full utilisation of contextual information and local information. Position attention modules are inserted in each stage between encoder and decoder to further enhance the spatial attention of extracted feature maps. Finally, a V-shaped decoder which is capable of utilising multi-level features is designed to get accurately dense prediction. The accuracy of urban scenes segmentation could significantly be enhanced in this way and successfully segmented the large scale variation objects from UAV views. Extensive ablation experiments and comparative experiments for the proposed network have been conducted on the public available urban scenes segmentation datasets for UAV imagery. Experimental results have demonstrated the effectiveness of designed network structure and the superiority of proposed network over state-of-the-art methods. Specifically, reached 53.2% mIoU on the UAVid dataset and 77.6% mIoU on the UDD6 dataset, respectively.

A Deep Learning Application to Map Weed Spatial Extent from Unmanned Aerial Vehicles Imagery

Weed infestation is a global threat to agricultural productivity, leading to low yields and financial losses. Weed detection, based on applying machine learning to imagery collected by Unmanned Aerial Vehicles (UAV) has shown potential in the past; however, validation on large data-sets (e.g., across a wide number of different fields) remains lacking, with few solutions actually made operational. Here, we demonstrate the feasibility of automatically detecting weeds in winter wheat fields based on deep learning methods applied to UAV data at scale. Focusing on black-grass (the most pernicious weed across northwest Europe), we show high performance (i.e., accuracy above 0.9) and highly statistically significant correlation (i.e., ro > 0.75 and p < 0.00001) between imagery-derived local and global weed maps and out-of-bag field survey data, collected by experts over 31 fields (205 hectares) in the UK. We demonstrate how the developed deep learning model can be made available via an easy-to-use docker container, with results accessible through an interactive dashboard. Using this approach, clickable weed maps can be created and deployed rapidly, allowing the user to explore actual model predictions for each field. This shows the potential for this approach to be used operationally and influence agronomic decision-making in the real world.

Predicting Sugarcane Biometric Parameters by UAV Multispectral Images and Machine Learning

Multispectral sensors onboard unmanned aerial vehicles (UAV) have proven accurate and fast to predict sugarcane yield. However, challenges to a reliable approach still exist. In this study, we propose to predict sugarcane biometric parameters by using machine learning (ML) algorithms and multitemporal data through the analysis of multispectral images from UAV onboard sensors. The research was conducted on five varieties of sugarcane, as a way to make a robust approach. Multispectral images were collected every 40 days and the evaluated biometric parameters were: number of tillers (NT), plant height (PH), and stalk diameter (SD). Two ML models were used: multiple linear regression (MLR) and random forest (RF). The results showed that models for predicting sugarcane NT, PH, and SD using time series and ML algorithms had accurate and precise predictions. Blue, Green, and NIR spectral bands provided the best performance in predicting sugarcane biometric attributes. These findings expand the possibilities for using multispectral UAV imagery in predicting sugarcane yield, particularly by including biophysical parameters.

Assessing Structural Complexity of Individual Scots Pine Trees by Comparing Terrestrial Laser Scanning and Photogrammetric Point Clouds

Structural complexity of trees is related to various ecological processes and ecosystem services. To support management for complexity, there is a need to assess the level of structural complexity objectively. The fractal-based box dimension (Db) provides a holistic measure of the structural complexity of individual trees. This study aimed to compare the structural complexity of Scots pine (Pinus sylvestris L.) trees assessed with Db that was generated with point cloud data from terrestrial laser scanning (TLS) and aerial imagery acquired with an unmanned aerial vehicle (UAV). UAV imagery was converted into point clouds with structure from motion (SfM) and dense matching techniques. TLS and UAV measured Db-values were found to differ from each other significantly (TLS: 1.51 ± 0.11, UAV: 1.59 ± 0.15). UAV measured Db-values were 5% higher, and the range was wider (TLS: 0.81–1.81, UAV: 0.23–1.88). The divergence between TLS and UAV measurements was found to be explained by the differences in the number and distribution of the points and the differences in the estimated tree heights and number of boxes in the Db-method. The average point density was 15 times higher with TLS than with UAV (TLS: 494,000, UAV 32,000 points/tree), and TLS received more points below the midpoint of tree heights (65% below, 35% above), while UAV did the opposite (22% below, 78% above). Compared to the field measurements, UAV underestimated tree heights more than TLS (TLS: 34 cm, UAV: 54 cm), resulting in more boxes of Db-method being needed (4–64%, depending on the box size). Forest structure (two thinning intensities, three thinning types, and a control group) significantly affected the variation of both TLS and UAV measured Db-values. Still, the divergence between the two approaches remained in all treatments. However, TLS and UAV measured Db-values were consistent, and the correlation between them was 75%.

Automated extraction of Camellia oleifera crown using unmanned aerial vehicle visible images and the ResU-Net deep learning model.

As one of the four most important woody oil-tree in the world, Camellia oleifera has significant economic value. Rapid and accurate acquisition of C. oleifera tree-crown information is essential for enhancing the effectiveness of C. oleifera tree management and accurately predicting fruit yield. This study is the first of its kind to explore training the ResU-Net model with UAV (unmanned aerial vehicle) images containing elevation information for automatically detecting tree crowns and estimating crown width (CW) and crown projection area (CPA) to rapidly extract tree-crown information. A Phantom 4 RTK UAV was utilized to acquire high-resolution images of the research site. Using UAV imagery, the tree crown was manually delineated. ResU-Net model’s training dataset was compiled using six distinct band combinations of UAV imagery containing elevation information [RGB (red, green, and blue), RGB-CHM (canopy height model), RGB-DSM (digital surface model), EXG (excess green index), EXG-CHM, and EXG-DSM]. As a test set, images with UAV-based CW and CPA reference values were used to assess model performance. With the RGB-CHM combination, ResU-Net achieved superior performance. Individual tree-crown detection was remarkably accurate (Precision = 88.73%, Recall = 80.43%, and F1score = 84.68%). The estimated CW (R2 = 0.9271, RMSE = 0.1282 m, rRMSE = 6.47%) and CPA (R2 = 0.9498, RMSE = 0.2675 m2, rRMSE = 9.39%) values were highly correlated with the UAV-based reference values. The results demonstrate that the input image containing a CHM achieves more accurate crown delineation than an image containing a DSM. The accuracy and efficacy of ResU-Net in extracting C. oleifera tree-crown information have great potential for application in non-wood forests precision management.

Semi-automated detection of ungulates using UAV imagery and reflective spectrometry

In the field of species conservation, the use of unmanned aerial vehicles (UAV) is increasing in popularity as wildlife observation and monitoring tools. With large datasets created by UAV-based species surveying, the need arose to automate the detection process of the species. Although the use of computer learning algorithms for wildlife detection from UAV-derived imagery is an increasing trend, it depends on a large amount of imagery of the species to train the object detector effectively. However, there are alternatives like object-based image analysis (OBIA) software available if a large amount of imagery of the species is not available to develop a computer-learned object detector.The study tested the semi-automated detection of reintroduced Arabian Oryx (O. leucoryx), using the specie's coat sRGB-colour profiles as input for OBIA to identify adult O. leucoryx, applied to UAV acquired imagery. Our method uses lab-measured spectral reflection of hair sample values, collected from captive O. leucoryx as an input for OBIA ruleset to identify adult O. leucoryx from UAV survey imagery using semi-automated supervised classification.The converted mean CIE Lab reflective spectrometry colour values of n = 50 hair samples of adult O. leucoryx to 8-bit sRGB-colour profiles of the species resulted in the red-band value of 157.450, the green-band value of 151.390 and blue-band value of 140.832. The sRGB values and a minimum size permitter were added as the input of the OBIA ruleset identified adult O. leucoryx with a high degree of efficiency when applied to three UAV census datasets.Using species sRGB-colour profiles to identify re-introduced O. leucoryx and extract location data using a non-invasive UAV-based tool is a novel method with enormous application possibilities. Coat refection sRGB-colour profiles can be developed for a range of species and customised to autodetect and classify the species from remote sensing data.

High-spatial resolution UAV multispectral data complementing satellite imagery to characterize a chinstrap penguin colony ecosystem on deception island (Antarctica)

ABSTRACT Remote sensing has evolved as an alternative to traditional techniques in the spatio-temporal monitoring of the Antarctic ecosystem, especially with the rapid expansion of the use of Unmanned Aerial Vehicles (UAVs), providing a centimeter-scale spatial resolution. In this study, the potential of a high-spatial resolution multispectral sensor embedded in a UAV is compared to medium resolution satellite remote sensing (Sentinel-2 and Landsat 8) to monitor the characteristics of the Vapor Col Chinstrap penguin (Pygoscelis antarcticus) colony ecosystem (Deception Island, South Shetlands Islands, Antarctica). Our main objective is to generate precise thematic maps of the typical ecosystem of penguin colonies derived from the supervised analysis of the spectral information obtained with these remote sensors. For this, two parametric classification algorithms (Maximum Likelihood, MLC, and Spectral Angle, SAC) and two non-parametric machine learning classifiers (Support Vector Machine, SVM, and Random Forest, RFC) are tested with UAV imagery, obtaining the best results with the SVM classifier (93.19% OA). Our study shows that the use of UAV outperforms satellite imagery (87.26% OA with Sentinel-2 Level 2 (S2L2) and 70.77% OA with Landsat 8 Level 2 (L8L2) in SVM classification) in the characterization of the substrate due to a higher spatial resolution, although differences between UAV and S2L2 are minimal. Thus, both sensors used in tandem could provide a broader and more precise view of how the area covered by the different elements of these ecosystems can change over time in a global climate change scenario. In addition, this study represents a precise UAV monitoring that takes place in this Chinstrap penguin colony, estimating a total coverage of approximately 20,000 m2 of guano areas in the study period.

Engineering geomorphological and InSAR investigation of an urban landslide, Gisborne, New Zealand

The East Coast region of New Zealand has some of the highest erosion rates in the world, due to its proximity to an active plate boundary, susceptibility to high-intensity storms and steep terrain underlain by young soft sedimentary rock and soil. In the city of Gisborne, expansion of residential blocks into steeper terrain in peri-urban areas has required improved capacity for the characterisation and monitoring of slope stability. Landslides have affected several properties and have destroyed infrastructure. Slope failure commonly occurs during heavy rainfall events when slow-moving retrogressive slides transition into earthflows and mudflows. In this study, we used in situ sampling and testing methods combined with remote sensing techniques to provide an understanding of the pre-failure and post-failure behaviour of an urban landslide in Gisborne. High-resolution aerial imagery, unmanned aerial vehicle imagery and light detection and ranging data revealed slope morphology and contours of prehistoric failures in the area, and highlighted the more recent impacts of deforestation on slope stability. Furthermore, Sentinel-1 InSAR analysis determined that gradual deformation began in 2017, following two ex-tropical cyclone events. Deformation downslope continued until an initial failure in July 2020. Following that event, some parts of the slope proceeded to accelerate, leading to a further reactivation event in November 2021, following heavy rainfall. During this November 2021 event, average line of sight velocities ranged from −7.9 to −11.2 mm/year, and deformation rates in the vertical direction (related to rotational slumping) averaged −11.2 to −11.9 mm/year, consistent with field observations.

Automatic detection of snow breakage at single tree level using YOLOv5 applied to UAV imagery

The assessment of forest abiotic damages such as snow breakage is important to ensure compensation to forest owners. Currently, information on the extent of snow breakage is gathered through time-consuming and potentially biased field surveys. In such situations where field surveys are still common practice, unmanned aerial vehicles (UAVs) are increasingly being used to provide a more cost-efficient and objective methods to answer forest information needs. Further, the advent of sophisticated computer vision techniques such as convolutional neural networks (CNNs) offers new ways to analyze image data more efficiently and accurately. We proposed an object detection method to automatically identify trees and classify them according to the damage by snow based on a YOLO CNN architecture. UAV imagery collected across 89 study areas and over the course of the entire year were manually annotated into a total of >55 K single trees classified as healthy, damaged, or dead. The annotated trees, along with the corresponding UAV imagery were used to train a YOLOv5 object detection model. Furthermore, we tested the effect of seasonality, and varying atmospheric and lighting conditions on the model’s performance. Based on an independent test set of data we found that the general model including all of the data (i.e. any seasons, atmospheric conditions, and time of the day) outperformed all other tested scenarios (i.e. precision = 62 %; recall = 61 %). Furthermore, we found that despite the fact that the snow damaged trees represented a minority class (i.e. 16 % of the annotated trees), they were detected with the largest precision (76 %) and recall (78 %). Finally, the general model transferred well across the variation in seasons, atmospheric and illumination conditions, making it suitable for usage for any new UAV image acquisition.

Mapping of Rumex obtusifolius in nature conservation areas using very high resolution UAV imagery and deep learning

Rumex obtusifolius (Rumex or broad leaved dock) is one of the most common weeds in grasslands. It spreads quickly, lowers the nutritional value of the grass, and is poisonous for livestock due to its oxalic acid content. Mapping it is important before any control treatment is applied. Current methods for mapping Rumex either involve manual work or the utilization of ground robots, which are not efficient in large fields. This study investigated the feasibility of using aerial images from unmanned aerial vehicles (UAV) and deep learning to map Rumex in grasslands. Seven pre-trained CNN models were tested using transfer learning on UAV images acquired at 10 m, 15 m, and 30 m height. Based on Cross Validation results, MobileNet performed the best in detecting Rumex, with an F1-Score of 78.36% and an AUROC of 93.74%, at 10 m height. At 15 m, the detection performance was relatively lower (F1-score = 72.00%, AUROC = 88.67%), but the results showed that the performance can increase with more data. Experiments also showed that Rumex detection was dependent on the flight height since the algorithm was unable to detect the plants at 30 m height. The code and the datasets used in this work were released in an open access repository to contribute to the advances in grassland management using UAV technology.

Remote Sensing Prescription for Rice Nitrogen Fertilizer Recommendation Based on Improved NFOA Model

Precise fertilization of rice depends on the timely and effective acquisition of fertilizer application recommended by prescription maps in large-scale cropland, which can provide fertilization spatial information reference. In this paper, the prescription map was discussed based on the improved nitrogen fertilizer optimization algorithm (NFOA), using satellite and unmanned aerial vehicle (UAV) imagery, and supplemented by meteorological data. Based on the principles of NFOA, firstly, remote sensing data and meteorological data were collected from 2019 to 2021 to construct a prediction model for the potential yield of rice based on the in-season estimated yield index (INSEY). Secondly, based on remote sensing vegetation indices (VIs) and spectral features of bands, the grain nitrogen content (GNC) prediction model constructed using the Random Forest (RF) algorithm was used to improve the values of GNC taken in the NFOA. The nitrogen demand for rice was calculated according to the improved NFOA. Finally, the nitrogen fertilizer application recommended prescription map of rice in large-scale cropland was generated based on UAV multispectral images, and the economic cost-effectiveness of the prescription map was analyzed. The analysis results showed that the potential yield prediction model of rice based on the improved INSEY had a high fitting accuracy (R2 = 0.62). The accuracy of GNC estimated with the RF algorithm reached 96.3% (RMSE = 0.07). The study shows that, compared with the non-directional and non-quantitative conventional tracking of N fertilizer, the recommended prescription map based on the improved NFOA algorithm in large-scale cropland can provide accurate information for crop N fertilizer variable tracking and provide effective positive references for the economic benefits of rice and ecological benefits of the field environment.

Kelp Patch-Specific Characteristics Limit Detection Capability of Rapid Survey Method for Determining Canopy Biomass Using an Unmanned Aerial Vehicle

Kelp forests dominate autotrophic biomass and primary productivity of approximately 30,000 to 60,000 km of shallow temperate and Arctic rocky reef coastline globally and contribute significantly to carbon cycling in the coastal ocean. Rapid biomass turnover is driven by very high growth rates and seasonal environmental drivers. As a result, kelp biomass varies greatly with time, space, and by species. In the northeast Pacific region, bull kelp (Nereocystis leutkeana) and giant kelp (Macrocystis pyrifera) form extensive floating surface canopies with a distinct spectral signature compared to the surrounding water. Studies have shown that remote sensing is advantageous for deriving large-scale estimates of floating surface canopy biomass, which comprises more than 90% of bull and giant kelp standing stock. However, development and validation of remotely derived kelp canopy biomass is lacking because existing approaches are time intensive and costly. This study attempted to close that gap by developing a rapid survey design utilizing diver and unmanned aerial vehicle (UAV) imagery across six sites in northern and central California. Kelp sporophytes were collected and measured for morphometric characteristics and genera-specific allometry to canopy biomass. Kelp density was measured using in situ diver surveys and coupled with UAV imagery to quantify kelp canopy biomass at a range of ground sampling distances. We successfully estimated kelp canopy biomass from UAV imagery at 33% (2/6) of the survey sites, but consistently determining canopy biomass via this approach was challenged by both survey design and kelp patch-specific spatial characteristics. The morphologies of bull kelp in Monterey were significantly different than other regions measured, but further work should be conducted to fully characterize differences in canopy biomass at the regional and sub-regional scale. We use this opportunity to suggest survey design strategies that will increase the success of future methodological development of UAV biomass retrieval. We also recommend developing long-term, annual genera-specific monitoring programs across the northeast Pacific region and beyond to validate remote sensing derived biomass estimates beyond the small number of existing well-characterized sites.

Fine-Scale Mapping of Soil Organic Matter in Agricultural Soils Using UAVs and Machine Learning

The fine-scale mapping of soil organic matter (SOM) in croplands is vital for the sustainable management of soil. Traditionally, SOM mapping relies on laboratory methods that are labor-intensive and costly. Recent advances in unmanned aerial vehicles (UAVs) afford new opportunities for rapid and low-cost SOM mapping at the field scale. However, the conversion from UAV measurements to SOM maps requires specific transfer models that still rely on local sampling. This study aimed to develop a method for predicting topsoil SOM at a high resolution on the field scale based on soil color information gained from low-altitude UAV imagery and machine learning. For this, we performed a UAV survey in cropland within the German loess belt. We used two fields, one for training and one for validation of the model, to test the model transferability. We analyzed 91 soil samples for SOM in the laboratory for the model calibration and 8 additional samples for external model validation. A random forest model (RF) showed good performance for the prediction of SOM based on UAV-derived color information with an RMSE of 0.13% and with an RPIQ of 2.42. The RF model was used to predict SOM at a point-support of 1 × 1 m. The SOM map revealed spatial patterns within the fields with a uniform spread of the prediction uncertainty. The validation of the model performed similarly to the calibration with an RMSE of 0.12% and an RPIQ of 2.05, albeit with a slight bias of 0.05%. This validation using external data showed that prediction models are transferable to neighboring fields, thus permitting the prediction on larger scale farms or enabling carbon monitoring over time.