Using Unmanned Aerial Vehicle Images Research Articles

Mapping large-scale pine wilt disease trees with a lightweight deep-learning model and very high-resolution UAV images

ABSTRACT Pine wilt disease (PWD), caused by pine wood nematodes, has brought a great loss in ecology and economy all over the world. In China, the forest health status is also significantly affected by PWD since 1980, especially in coniferous forests and mixed forest regions. The PWD spreads very fast and can cause a healthy pine wood tree to die within a very short time. An effective way to protect other healthy pine wood trees is to discover PWD trees early. Using unmanned aerial vehicle (UAV) images can help people discover the PWD tree quickly and accurately, and a few automatical methods have been developed to monitor PWD trees including the deep learning methods. Because of the robust spatial-temporal transferability, deep learning methods have become the mainstream algorithms to monitor PWD trees. As we know, the training dataset is the most important material to train a deep learning model. However, there is still a lack of a PWD segmentation dataset so far. To fill this gap, in this paper, we have generated the first open-sourced PWD segmentation dataset based on very high-resolution UAV images to help the community conduct PWD monitor research conveniently. This dataset has 994 training samples, and each sample has visible bands with 512 × 512 pixels, and the spatial resolution of this dataset is 0.05 m. In order to train an advanced segmentation model, we have designed a lightweight deep-learning model for mobile devices or edge devices in this manuscript, named MobileSeg. The main feature of MobileSeg is its decoupling, which uses the re-parameterization technology to improve the model performance. Finally, a large-scale real-world scenario experiment with very high-resolution UAV images was utilized to validate the performance of MobileSeg, the experiment result indicated that MobileSeg has achieved the best performance compared with the recent lightweight segmentation models, and the experiment also proved the effectiveness of the proposed PWD dataset.

3D Model of Tunku Tun Aminah Library Using Unmanned Aerial Vehicle (UAV)

3-Dimensional (3D) modelling helped bring flat, 2D ideas to life. 3D models were valuable because they could include a much wider array of project information. The purpose of this study was to develop a 3D model by using Unmanned Aerial Vehicle (UAV) images of Tunku Tun Aminah (PTTA) Library building. To create a 3D model, Agisoft Metashape software combined photos from UAV images. Then, we assessed the accuracy of the 3D model of Tunku Tun Aminah's library building to the actual scale of the building from a vertical and horizontal view. The result of the 3D model that has been constructed demonstratesa high level of accuracy in capturing its intricate details and spatial characteristics. The vertical and horizontal views, both the Tunku Tun Aminah library and the parking lots hadactual lengths that matched the estimated values precisely. Theresults indicated that there were slight differences in measurements, and all the measurements were 95% accurate and above. The precision demonstrated in capturing the dimensions of these structures indicates reliable data for analysis and decision-making purposes. The Ground Sampling Distance (GSD) for the assessment was 1.36 cm per pixel. At the end of this research, the 3D model of PTTA library was uploaded on SketchFab to allow various parties, including university authorities, contractors, local authorities, architects, and structural engineers to have a better understanding of the building’s condition and its surroundings.

Spoil characterisation using UAV-based optical remote sensing in coal mine dumps

The structural integrity of mine dumps is crucial for mining operations to avoid adverse impacts on the triple bottom-line. Routine temporal assessments of coal mine dumps are a compliant requirement to ensure design reconciliation as spoil offloading continues over time. Generally, the conventional in-situ coal spoil characterisation is inefficient, laborious, hazardous, and prone to experts' observation biases. To this end, this study explores a novel approach to develop automated coal spoil characterisation using unmanned aerial vehicle (UAV) based optical remote sensing. The textural and spectral properties of the high-resolution UAV images were utilised to derive lithology and geotechnical parameters (i.e., fabric structure and relative density/consistency) in the proposed workflow. The raw images were converted to an orthomosaic using structure from motion aided processing. Then, structural descriptors were computed per pixel to enhance feature modalities of the spoil materials. Finally, machine learning algorithms were employed with ground truth from experts as training and testing data to characterise spoil rapidly with minimal human intervention. The characterisation accuracies achieved from the proposed approach manifest a digital solution to address the limitations in the conventional characterisation approach.

Estimation of canopy nitrogen nutrient status in lodging maize using unmanned aerial vehicles hyperspectral data

Rapid and non-destructive monitoring of the temporal dynamic changes in canopy nitrogen in lodging maize is essential to explore the influence on plant nutrient transport and yield loss. This study aims to monitor the canopy nitrogen status and lodging severity in maize using unmanned aerial vehicle (UAV) hyperspectral technology. The lodging maize experiments, involving different types, were conducted at the vegetative tasseling (VT) and reproductive milk (R3) stages. Agronomic traits and hyperspectral images were collected at 1, 7, 14, 21 and 28 days after lodging (DAL). Lodging intensity was quantified using canopy nitrogen concentration (CNC) and canopy nitrogen volumetric density (CNVD). Firstly, the variation in CNC and CNVD among different lodging types was analyzed, and the correlation between CNC, CNVD and the canopy hyperspectrum were assessed. Subsequently, the recursive feature elimination with cross-validation (RFECV) algorithm was used to screen wavelengths sensitive to CNC and CNVD. Next, the CNC and CNVD estimation models were developed using random forest regression (RFR) and gradient boosting regression (GBR) algorithms. Finally, the spatial distribution maps of CNC and CNVD in lodging maize were generated based on UAV images. The key findings were as follows: (1) CNVD exhibited more significant variation than CNC in different lodging types, with greater lodging severity corresponding to higher CNVD values; (2) the correlation between CNVD and canopy hyperspectrum was stronger than that of CNC; (3) there were 12 CNC-sensitive and 16 CNVD-sensitive wavelengths selected by RFECV algorithm; (4) the CNVD estimation model performed the best (R2 = 0.77, RMSE = 69.61 g/m3 in testing set) using the GBR algorithm. Therefore, the combination of feature selection with regression models effectively reduces hyperspectral data dimensionality. This enhances the estimation accuracy and computational efficiency for assessing canopy nitrogen nutrient status in lodging maize.

Assessing Within-Field Variation in Alfalfa Leaf Area Index Using UAV Visible Vegetation Indices

This study examines the use of leaf area index (LAI) to inform variable-rate irrigation (VRI) for irrigated alfalfa (Medicago sativa). LAI is useful for predicting zone-specific evapotranspiration (ETc). One approach toward estimating LAI is to utilize the relationship between LAI and visible vegetation indices (VVIs) using unmanned aerial vehicle (UAV) imagery. This research has three objectives: (1) to measure and describe the within-field variation in LAI and canopy height for an irrigated alfalfa field, (2) to evaluate the relationships between the alfalfa LAI and various VVIs with and without field average canopy height, and (3) to use UAV images and field average canopy height to describe the within-field variation in LAI and the potential application to VRI. The study was conducted in 2021–2022 in Rexburg, Idaho. Over the course of the study, the measured LAI varied from 0.23 m2 m−2 to 11.28 m2 m−2 and canopy height varied from 6 cm to 65 cm. There was strong spatial clustering in the measured LAI but the spatial patterns were dynamic between dates. Among eleven VVIs evaluated, the four that combined green and red wavelengths but excluded blue wavelengths showed the most promise. For all VVIs, adding average canopy height to multiple linear regression improved LAI prediction. The regression model using the modified green–red vegetation index (MGRVI) and canopy height (R2 = 0.93) was applied to describe the spatial variation in the LAI among VRI zones. There were significant (p < 0.05) but not practical differences (<15%) between pre-defined zones. UAV imagery coupled with field average canopy height can be a useful tool for predicting LAI in alfalfa.

Geometric consistency enhanced deep convolutional encoder-decoder for urban seismic damage assessment by UAV images

Precise and rapid assessment of seismic damage to buildings is critical for urban regions. To address this challenge, this study proposes QuakeCityNet (QCNet-M-N)- a model with flexible configurations of M encoding stages and N embedding convolution operations for exact pixel-level recognition of earthquake-damaged buildings using unmanned aerial vehicle (UAV) images. A novel loss function, geometric consistency enhanced (GCE) loss, is designed to focus on the building regions and local boundaries, taking into account the geometrical constraints of split line length, curvature, and area. Test results indicate that the proposed QCNet model can achieve robust and stable segmentation accuracy under diverse weather conditions, such as abnormal illumination, rain, and fog. Moreover, the utilization of GCE loss significantly reduces the false-positive small-region noise while preserving overall geometrical shapes. Finally, an application of seismic assessment is conducted in Beichuan county to demonstrate the effectiveness of the proposed method.

Automated Road Damage Detection Using UAV Images and Deep Learning Techniques

This paper presents a novel automated road damage detection approach using Unmanned Aerial Vehicle (UAV) images and deep learning techniques. Maintaining road infrastructure is critical for ensuring a safe and sustainable transportation system. However, the manual collection of road damage data can be labor-intensive and unsafe for humans. Therefore, we propose using UAVs and Artificial Intelligence (AI) technologies to improve road damage detection’s efficiency and accuracy significantly. Our proposed approach utilizes three algorithms, YOLOv4, YOLOv5, and YOLOv7, for object detection and localization in UAV images. We trained and tested these algorithms using a combination of the RDD2022 dataset from China and a Spanish road dataset. The experimental results demonstrate that our approach is efficient and achieves 59.9% mean average precision mAP@.5 for the YOLOv5 version, 73.20% mAP@.5 for the YOLOv7 version, and 65.70% mAP@.5 for a YOLOv5 model with a Transformer Prediction Head. These results demonstrate the potential of using UAVs and deep learning for automated road damage detection and pave the way for future research in this field.

IUNet-IF: identification of construction waste using unmanned aerial vehicle remote sensing and multi-layer deep learning methods

ABSTRACT Considering current problems in the identifying, locating, and supervising of construction waste, it is difficult to achieve real-time target judgement and identification in the short term. This study proposes a deep learning-based construction waste identification and classification system, named IUNet-IF, developed using unmanned aerial vehicle (UAV) images. High-resolution UAV aerial images were used to establish a sample database and train U-Net models. The U-Net model, which exhibited the highest overall performance and identification accuracy, was enhanced to develop an IUNet model. Subsequently, the texture and colour features of UAV images were extracted and input to the IUNet model to build a novel IUNet-IF model. The proposed method uses the optimal model and an input combination of texture features, colour features, and original images, which improves the segmentation results compared to those when only the original image is used. The experimental results indicate that the system can achieve a maximum pixel accuracy index, mean intersection over union (mIoU), and mean pixel accuracy (MPA) of 98.98%, 93.78%, and 97.26%, respectively. These results verify the feasibility of extracting construction waste covered by a dust-proof net from UAV remote sensing images and can provide reference information for the automatic extraction of construction waste.

A novel strategy for pest disease detection of Brassica chinensis based on UAV imagery and deep learning

ABSTRACT Pest disease is considered to be among the greatest threats to Brassica chinensis. The physical invasion caused by pests will form wormholes, which seriously affect the yield and quality of Brassica chinensis. To tackle this problem, we innovatively propose automatic pest disease detection by taking images using unmanned aerial vehicle (UAV). Compared to the traditional on-site manual inspection and fixed camera-based image detection methods, it has the advantages of high efficiency, low cost, and suitable for a wide range of scenes. However, pest disease detection through UAV images faces many challenges, such as image blur and small object size, and it is difficult for previous methods to achieve accurate and effective detection. Therefore, we propose a strategy for pest disease detection of Brassica chinensis in aerial images based on deep learning. At first, we employed an image tiling module to preprocess the aerial images. Then, an image restoration module is adopted to deblur the image. Third, an advanced method of disease object detection based on improved CenterNet that combines attention mechanism and DIoU loss is proposed. In addition, a dataset of aerial images of Brassica chinensis bitten by flea beetle was built and annotated, containing 42,452 wormhole annotations. Extensive comparative experimental results show that the detection performance of the proposed method is superior to the existing detection methods. And we achieved an overall accuracy of 87.2% AP50 and 94.7% R-squared for pest disease detection of Brassica chinensis, which can meet the requirements of real application.

A Wildfire Smoke Detection System Using Unmanned Aerial Vehicle Images Based on the Optimized YOLOv5

Wildfire is one of the most significant dangers and the most serious natural catastrophe, endangering forest resources, animal life, and the human economy. Recent years have witnessed a rise in wildfire incidents. The two main factors are persistent human interference with the natural environment and global warming. Early detection of fire ignition from initial smoke can help firefighters react to such blazes before they become difficult to handle. Previous deep-learning approaches for wildfire smoke detection have been hampered by small or untrustworthy datasets, making it challenging to extrapolate the performances to real-world scenarios. In this study, we propose an early wildfire smoke detection system using unmanned aerial vehicle (UAV) images based on an improved YOLOv5. First, we curated a 6000-wildfire image dataset using existing UAV images. Second, we optimized the anchor box clustering using the K-mean++ technique to reduce classification errors. Then, we improved the network's backbone using a spatial pyramid pooling fast-plus layer to concentrate small-sized wildfire smoke regions. Third, a bidirectional feature pyramid network was applied to obtain a more accessible and faster multi-scale feature fusion. Finally, network pruning and transfer learning approaches were implemented to refine the network architecture and detection speed, and correctly identify small-scale wildfire smoke areas. The experimental results proved that the proposed method achieved an average precision of 73.6% and outperformed other one- and two-stage object detectors on a custom image dataset.

Deep convolutional neural networks and Swin transformer-based frameworks for individual date palm tree detection and mapping from large-scale UAV images

Timely and reliable mapping of individual date palm trees is essential for their monitoring, health and risk assessment, pest control, and sustainable management of the date palm industry. This study presents an instance segmentation framework for large-scale detection and mapping of date palm trees using unmanned aerial vehicle (UAV)-based images. First, a data conversion framework is created to convert UAV image tiles and ground-truth vector data into annotation format of Common Objects in Context. Second, this study examines the efficacy of various instance segmentation models, namely, mask region convolutional neural network (Mask R-CNN), Mask Scoring R-CNN, You Only Look At CoefficientTs, Point-based Rendering, Segmenting Objects by Locations (SOLO), and SOLOv2) with varying residual learning networks (ResNets) in detecting and delineating individual date palm trees. Furthermore, the performance of two variants of Swin Transformer networks with a feature pyramid network (FPN) (Swin-small-FPN and Swin-tiny-FPN) as Mask R-CNN network backbones was also evaluated. Third, we assess the generalizability of the evaluated instance segmentation models and backbones on different testing datasets with varying spatial resolutions. Results show that Mask R-CNN models based on Swin Transformers backbones outperform those with ResNets in the detection and segmentation of date palm trees with mAP50 of 92% and 91% and F-measures of 94% and 93%. Moreover, the Mask scoring R-CNN-based ResNet-50 and Mask R-CNN with a Swin-small-FPN backbone outperform the evaluated models and demonstrate great generalizability in different datasets with diverse spatial resolutions. The proposed instance segmentation framework provides an efficient tool for date palm tree mapping from multi-scale UAV-based images and is valuable and suitable for individual tree crown delineations and other earth-related applications.

Origin and Abundance of Steller Sea Lions (Eumetopias jubatus) in Winter Haulout at Benten-Jima Rock Off Cape Soya, Hokkaido, Japan between 2012–2017

Benten-Jima Rock, located off Cape Soya, Hokkaido, has been a Steller sea lion (SSL; Eumetopias jubatus) winter haulout for decades. The animals usually occupy the site from October to May. Observations have been sporadic, although the population count started to increase in 2005. We have monitored SSL numbers since 2012 using several survey methods, such as observation by direct counting and remote archival cameras. Since these data were not sufficient, owing to blind spots, we started using unmanned aerial vehicle (UAV) surveys in 2016 and corrected the previously collected data. Using these methods, a considerable number of SSLs were observed at Benten-Jima Rock during 2016–2017. The maximum number of SSLs was 3158 on land and 3056 in the water near the site, as counted from UAV images on May 2, 2017. Based on hot brand marks, we found that Benten-Jima Rock hosted SSLs from all ten main rookeries along the Asian coast. The majority (∼60%) were from Tuleny Island near the east coast of Sakhalin. The cause of this extraordinary increase in SSL numbers at Benten-Jima Rock remains unclear and requires further monitoring and research.

Generation of Dense and High-Precision Digital Elevation Model Using Low-Cost Unmanned Aerial Vehicle and Space-Borne TanDEM-X to Measure Exposed Area Change Due to Tidal Invasion

Tidal flats are internationally protected areas because of their environmental, geological, and economic value. Nevertheless, these are areas where casualties are common among visitors. To ensure the safety of visitors, efforts must be made to understand the geomorphologic characteristics of tidal flats. It is necessary to use three-dimensional topographic data to clearly determine when an exposed tidal flat rapidly disappears as a result of periodic tidal fluctuations. Digital elevation models (DEMs) created using unmanned aerial vehicle (UAV) imaging and commercial photogrammetric software to measure tidal flats are produced using ground control points (GCPs). It is difficult to conduct field surveys and readings of image coordinates that correspond to these because tidal flat areas are difficult to access. The distribution of GCPs affects the accuracy of a DEM because the entire test region cannot be covered during a survey. Without GCPs, the DEM generated with the UAV images and photogrammetric software using the structure from motion (SfM) technique probably has nonlinear distortion such as a bowl shape. This paper proposes a practical method of least-squares three-dimensional (3D) surface matching with a polynomial model (LS3D-PM) to correct a distorted DEM. For this LS3D-PM matching, a global TanDEM-X DEM was used instead of a field survey to collect the GCPs in a tidal flat. Practical testing was conducted using the proposed method. The root mean square error with respect to the height decreased from approximately 1.0 m to 0.1 m, and the bowl-effect error was eliminated. With the DEM corrected using the proposed method and the classified orthoimages, the exposed tidal flat changes due to seawater invasion were estimated at different times during flood tide. This method has non-trivial technical values for quick and relatively precise DEM generations at a low cost compared to that of other techniques, particularly in emergency situations such as tsunamis, landslides, and earthquakes.

Road Traffic Monitoring from UAV Images Using Deep Learning Networks

In this paper, we propose a deep neural network-based method for estimating speed of vehicles on roads automatically from videos recorded using unmanned aerial vehicle (UAV). The proposed method includes the following; (1) detecting and tracking vehicles by analyzing the videos, (2) calculating the image scales using the distances between lanes on the roads, and (3) estimating the speeds of vehicles on the roads. Our method can automatically measure the speed of the vehicles from the only videos recorded using UAV without additional information in both directions on the roads simultaneously. In our experiments, we evaluate the performance of the proposed method with the visual data at four different locations. The proposed method shows 97.6% recall rate and 94.7% precision rate in detecting vehicles, and it shows error (root mean squared error) of 5.27 km/h in estimating the speeds of vehicles.

A New Method for Crop Row Detection Using Unmanned Aerial Vehicle Images

Crop row detection using unmanned aerial vehicle (UAV) images is very helpful for precision agriculture, enabling one to delineate site-specific management zones and to perform precision weeding. For crop row detection in UAV images, the commonly used Hough transform-based method is not sufficiently accurate. Thus, the purpose of this study is to design a new method for crop row detection in orthomosaic UAV images. For this purpose, nitrogen field experiments involving cotton and nitrogen and water field experiments involving wheat were conducted to create different scenarios for crop rows. During the peak square growth stage of cotton and the jointing growth stage of wheat, multispectral UAV images were acquired. Based on these data, a new crop detection method based on least squares fitting was proposed and compared with a Hough transform-based method that uses the same strategy to preprocess images. The crop row detection accuracy (CRDA) was used to evaluate the performance of the different methods. The results showed that the newly proposed method had CRDA values between 0.99 and 1.00 for different nitrogen levels of cotton and CRDA values between 0.66 and 0.82 for different nitrogen and water levels of wheat. In contrast, the Hough transform method had CRDA values between 0.93 and 0.98 for different nitrogen levels of cotton and CRDA values between 0.31 and 0.53 for different nitrogen and water levels of wheat. Thus, the newly proposed method outperforms the Hough transform method. An effective tool for crop row detection using orthomosaic UAV images is proposed herein.

A Study on Obtaining Tree Data from Green Spaces in Parks Using Unmanned Aerial Vehicle Images: Focusing on Mureung Park in Chuncheon

Background and objective The purpose of study is to analyze the three-dimensional (3D) structure by creating a 3D model for green spaces in a park using unmanned aerial vehicle (UAV) images. Methods After producing a digital surface model (DSM) and a digital terrain model (DTM) using UAV images taken in Mureung Park in Chuncheon-si, we generated a digital tree height model (DHM). In addition, we used the mean shift algorithm to test the classification accuracy, and obtain accurate tree height and volume measures through field survey. Results Most of the tree species planted in Mureung Park were Pinus koraiensis, followed by Pinus densiflora, and Zelkova serrata, and most of the shrubs planted were Rhododendron yedoense, followed by Buxus microphylla, and Spiraea prunifolia. The average height of trees measured at the site was 7.8 m, and the average height estimated by the model was 7.5 m, showing a difference of about 0.3 m. As a result of the t-test, there was no significant difference between height values of the field survey data and the model. The estimated green coverage and volume of the study site using the UAV were 5,019 m2 and 14,897 m3, respectively, and the green coverage and volume measured through the field survey were 6,339 m2 and 17,167 m3. It was analyzed that the green coverage showed a difference of about 21% and the volume showed a difference of about 13%. Conclusion The UAV equipped with RTK (Real-Time Kinematic) and GNSS (Global Navigation Satellite System) modules used in this study could collect information on tree height, green coverage, and volume with relatively high accuracy within a short period of time. This could serve as an alternative to overcome the limitations of time and cost in previous field surveys using remote sensing techniques.

Improving Potato Yield Prediction by Combining Cultivar Information and UAV Remote Sensing Data Using Machine Learning

Accurate high-resolution yield maps are essential for identifying spatial yield variability patterns, determining key factors influencing yield variability, and providing site-specific management insights in precision agriculture. Cultivar differences can significantly influence potato (Solanum tuberosum L.) tuber yield prediction using remote sensing technologies. The objective of this study was to improve potato yield prediction using unmanned aerial vehicle (UAV) remote sensing by incorporating cultivar information with machine learning methods. Small plot experiments involving different cultivars and nitrogen (N) rates were conducted in 2018 and 2019. UAV-based multi-spectral images were collected throughout the growing season. Machine learning models, i.e., random forest regression (RFR) and support vector regression (SVR), were used to combine different vegetation indices with cultivar information. It was found that UAV-based spectral data from the early growing season at the tuber initiation stage (late June) were more correlated with potato marketable yield than the spectral data from the later growing season at the tuber maturation stage. However, the best performing vegetation indices and the best timing for potato yield prediction varied with cultivars. The performance of the RFR and SVR models using only remote sensing data was unsatisfactory (R2 = 0.48–0.51 for validation) but was significantly improved when cultivar information was incorporated (R2 = 0.75–0.79 for validation). It is concluded that combining high spatial-resolution UAV images and cultivar information using machine learning algorithms can significantly improve potato yield prediction than methods without using cultivar information. More studies are needed to improve potato yield prediction using more detailed cultivar information, soil and landscape variables, and management information, as well as more advanced machine learning models.

Deep neural networks based automated extraction of dugong feeding trails from UAV images in the intertidal seagrass beds

Dugongs (Dugong dugon) are seagrass specialists distributed in shallow coastal waters in tropical and subtropical seas. The area and distribution of the dugongs’ feeding trails, which are unvegetated winding tracks left after feeding, have been used as an indicator of their feeding ground utilization. However, current ground-based measurements of these trails require a large amount of time and effort. Here, we developed effective methods to observe the dugongs’ feeding trails using unmanned aerial vehicle (UAV) images (1) by extracting the dugong feeding trails using deep neural networks. Furthermore, we demonstrated two applications as follows; (2) extraction of the daily new feeding trails with deep neural networks and (3) estimation the direction of the feeding trails. We obtained aerial photographs from the intertidal seagrass bed at Talibong Island, Trang Province, Thailand. The F1 scores, which are a measure of binary classification model’s accuracy taking false positives and false negatives into account, for the method (1) were 89.5% and 87.7% for the images with ground sampling resolutions of 1 cm/pixel and 0.5 cm/pixel, respectively, while the F1 score for the method (2) was 61.9%. The F1 score for the method (1) was high enough to perform scientific studies on the dugong. However, the method (2) should be improved, and there remains a need for manual correction. The mean area of the extracted daily new feeding trails from September 12–27, 2019, was 187.8 m2 per day (n = 9). Total 63.9% of the feeding trails was estimated to have direction within a range of 112.5° and 157.5°. These proposed new methods will reduce the time and efforts required for future feeding trail observations and contribute to future assessments of the dugongs’ seagrass habitat use.

Comparison of Outdoor Compost Pile Detection Using Unmanned Aerial Vehicle Images and Various Machine Learning Techniques

Since outdoor compost piles (OCPs) contain large amounts of nitrogen and phosphorus, they act as a major pollutant that deteriorates water quality, such as eutrophication and green algae, when the OCPs enter the river during rainfall. In South Korea, OCPs are frequently used, but there is a limitation that a lot of manpower and budget are consumed to investigate the current situation, so it is necessary to efficiently investigate the OCPs. This study compared the accuracy of various machine learning techniques for the efficient detection and management of outdoor compost piles (OCPs), a non-point pollution source in agricultural areas in South Korea, using unmanned aerial vehicle (UAV) images. RGB, multispectral, and thermal infrared UAV images were taken in August and October 2019. Additionally, vegetation indices (NDVI, NDRE, ENDVI, and GNDVI) and surface temperature were also considered. Four machine learning techniques, including support vector machine (SVM), decision tree (DT), random forest (RF), and k-NN, were implemented, and the machine learning technique with the highest accuracy was identified by adjusting several variables. The accuracy of all machine learning techniques was very high, reaching values of up to 0.96. Particularly, the accuracy of the RF method with the number of estimators set to 10 was highest, reaching 0.989 in August and 0.987 in October. The proposed method allows for the prediction of OCP location and area over large regions, thereby foregoing the need for OCP field measurements. Therefore, our findings provide highly useful data for the improvement of OCP management strategies and water quality.

ICENETv2: A Fine-Grained River Ice Semantic Segmentation Network Based on UAV Images

Accurate ice segmentation is one of the most crucial techniques for intelligent ice monitoring. Compared with ice segmentation, it can provide more information for ice situation analysis, change trend prediction, and so on. Therefore, the study of ice segmentation has important practical significance. In this study, we focused on fine-grained river ice segmentation using unmanned aerial vehicle (UAV) images. This has the following difficulties: (1) The scale of river ice varies greatly in different images and even in the same image; (2) the same kind of river ice differs greatly in color, shape, texture, size, and so on; and (3) the appearances of different kinds of river ice sometimes appear similar due to the complex formation and change procedure. Therefore, to perform this study, the NWPU_YRCC2 dataset was built, in which all UAV images were collected in the Ningxia–Inner Mongolia reach of the Yellow River. Then, a novel semantic segmentation method based on deep convolution neural network, named ICENETv2, is proposed. To achieve multiscale accurate prediction, we design a multilevel features fusion framework, in which multi-scale high-level semantic features and lower-level finer features are effectively fused. Additionally, a dual attention module is adopted to highlight distinguishable characteristics, and a learnable up-sampling strategy is further used to improve the segmentation accuracy of the details. Experiments show that ICENETv2 achieves the state-of-the-art on the NWPU_YRCC2 dataset. Finally, our ICENETv2 is also applied to solve a realistic problem, calculating drift ice cover density, which is one of the most important factors to predict the freeze-up data of the river. The results demonstrate that the performance of ICENETv2 meets the actual application demand.