Application of Deep-Learning Methods to Bird Detection Using Unmanned Aerial Vehicle Imagery

In order to ensure the safety of transmission lines, the use of unmanned aerial vehicle (UAV) images for automatic object detection has important application prospects, such as the detection of birds’ nests. The traditional bird’s nest detection methods mainly include the study of morphological characteristics of the bird’s nest. These methods have poor applicability and low accuracy. In this work, we propose a deep learning-based birds’ nests automatic detection framework—region of interest (ROI) mining faster region-based convolutional neural networks (RCNN). First, the prior dimensions of anchors are obtained by using k-means clustering to improve the accuracy of coordinate boxes generation. Second, in order to balance the number of foreground and background samples in the training process, the focal loss function is introduced in the region proposal network (RPN) classification stage. Finally, the ROI mining module is added to solve the class imbalance problem in the classification stage, combined with the characteristics of difficult-to-classify bird’s nest samples in the UAV images. After parameter optimization and experimental verification, the deep learning-based bird’s nest automatic detection framework proposed in this work achieves high detection accuracy. In addition, the mean average precision (mAP) and formula 1 (F1) score of the proposed method are higher than the original faster RCNN and cascade RCNN. Our comparative analysis verifies the effectiveness of the proposed method.

In potato (Solanum tuberosum) production, the number of tubers harvested and their sizes are related to the plant population. Field maps of the spatial variation in plant density can therefore provide a decision support tool for spatially variable harvest timing to optimize tuber sizes by allowing densely populated management zones more tuber-bulking time. Computer vision has been proposed to enumerate plant numbers using images from unmanned aerial vehicles (UAV) but inaccurate predictions in images of merged canopies remains a challenge. Some research has been done on individual potato plant bounding box prediction but there is currently no information on the spatial structure of plant density that these models may reveal and its relationship with potato yield quality attributes. In this study, the Faster Region-based Convolutional Neural Network (FRCNN) framework was used to produce a plant detection model and estimate plant densities across a UAV orthomosaic. Using aerial images of 2 mm ground sampling distance (GSD) collected from potatoes at 40 days after planting, the FRCNN model was trained to an average precision (aP) of 0.78 on unseen testing data. The model was then used to generate predictions on quadrants imposed on orthorectified rasters captured at 14 and 18 days after emergence. After spatially interpolating the plant densities, the resultant surfaces were highly correlated to manually-determined plant density (R2 = 0.80). Further correlations were observed with tuber number (r = 0.54 at Butter Hill; r = 0.53 at Horse Foxhole), marketable tuber weight per plant (r = −0.57 at Buttery Hill; r = −0.56 at Horse Foxhole) and the normalized difference vegetation index (r = 0.61). These results show that accurate two-dimensional maps of plant density can be constructed from UAV imagery with high correlation to important yield components, despite the loss of accuracy of FRCNN models in partially merged canopies.

The use of computer vision and deep learning in boundary documentation for land registration stems from the ongoing demand for appropriate mapping approaches of unregistered land rights to eradicate the global challenge of tenure insecurity. Previous research has yielded promising results towards automated extraction of photo‐visible cadastral boundaries from high‐resolution imagery. Nonetheless, the extraction of invisible cadastral boundaries is still a challenge. This study investigates the place of sensor/s on‐board unmanned aerial vehicles and deep learning algorithms in detecting cadastral boundaries. It develops a participatory boundary marking procedure using low‐cost markers to bring monument to previously invisible and ill‐defined cadastral boundaries. After that, the researchers trained and tested the accuracy of a convolutional neural network, namely single shot multi‐box detector (SSD) based on Residual Neural Network (ResNet) and Visual Geometry Group (VGG) backbone networks to automatically detect cadastral boundary markers from unmanned aerial vehicle imagery. SSD based on ResNet34 performed best with 0.88 precision, 0.92 recall and 0.91 F measure or (F1) score. VGG19‐based SSD yielded a precision of 0.47, recall of 0.53 and F1 score of 0.50. The horizontal accuracy of the cadastral map generated varied from 0.089 to 0.496 m per parcel, with a standard deviation of 0.120 m. Results show that this approach is practical for cadastral mapping in rural areas.

Modern people who value healthy eating habits have shown increasing interest in plum (Prunus mume) fruits, primarily owing to their nutritiousness and proven efficacy. As consumption increases, it becomes important to monitor work to prevent Prunus mume fruits from falling out. Moreover, determining the growth status of Prunus mume is also crucial and is attracting increasing attention. In this study, convolutional neural network (CNN)-based deep learning object detection was developed using RGBD images collected from Prunus mume farms. These RGBD images consider various environments, including the depth information of objects in the outdoor field. A faster region-based convolutional neural network (R-CNN), EfficientDet, Retinanet, and Single Shot Multibox Detector (SSD) were applied for detection, and the performance of all models was estimated by comparing their respective computing speeds and average precisions (APs). The test results show that the EfficientDet model is the most accurate, and SSD MobileNet is the fastest among the four models. In addition, the algorithm was developed to acquire the growth status of P. mume fruits by applying the coordinates and score values of bounding boxes to the depth map. Compared to the diameters of the artificial Prunus mume fruits used as the experimental group, the calculated diameters were very similar to those of the artificial objects. Collectively, the results demonstrate that the CNN-based deep learning Prunus mume detection and growth estimation method can be applied to real farmlands.

Real-Time and Accurate State Recognition of Latex Collection Bowls From Unmanned Aerial Vehicle Imagery Using an Improved You Only Look Once Detector With Multi-Scale Attention

Accurate occupancy information is crucial for enhancing energy efficiency and reducing carbon emissions in buildings. However, the inherent unpredictability of occupants introduces uncertainties in energy analysis and control strategy development. To address these challenges, this study proposes a vision-based method employing state-of-the-art deep learning models to capture real-time occupancy profiles in crowded indoor spaces. Utilising a self-collected image dataset, various deep learning models, including Single Shot MultiBox Detector (SSD), Faster Region-based Convolutional Neural Networks (Faster R-CNN), and different versions of You Only Look Once (YOLO) were trained and evaluated. An experiment was conducted in a lecture room equipped with cameras and environmental sensors to evaluate the performance of each model in terms of precision, computational efficiency, and adaptability to varying occupancy levels during a lecture session. The session included varying occupancy conditions: entering (barely occupied), during the lecture (typical occupancy), and leaving the room (again barely occupied). Among the models tested, YOLOv8x exhibited the best performance in terms of accuracy, while SSD lagged notably. The impact on the detection performance of various locations of camera setups was also explored. Energy simulations revealed that deep learning-based model generated occupancy profiles significantly deviated from conventional “fixed” occupancy profiles, resulting in a 13.45 % variation in predicted heating energy demand. However, compared to the ground truth, these profiles showed minimal variation (up to 6.72 %) for the Faster R-CNN and YOLO models, highlighting their accuracy and robustness. Additionally, although the deep learning-based occupancy profiles generally overpredicted the recorded data, the CO2 concentration trends they predicted aligned closely with the recorded data, unlike the “fixed” occupancy profiles. The findings underscore the importance of realistic occupancy profiles for reliable energy predictions in buildings and demonstrate the potential of the proposed vision-based method for advancing occupancy detection and building energy management.

WITHDRAWN: Effective utilization of YOLOv3 model for aircraft detection in Remotely Sensed Images

Hyperspectral imagery from an Unmanned Aerial Vehicle (UAV) has been used to predict soil organic carbon (SOC) content. This study compared UAV hyperspectral imagery with Sentinel-2A multispectral imagery and soil spectra in the laboratory to predict the SOC content in Suaeda salsa (S. salsa) coastal wetland. Characteristic variables were determined through correlation analysis comparing observed SOC with constructed spectral indices of SOC, index variables, and spectral bands. Predict Models of SOC were constructed by multilayer perceptron (MLP), the prediction accuracy was evaluated and the spatial maps of SOC were produced. The results showed that estimating SOC using UAV, Sentinel imagery, and laboratory spectra has great potential when using MLP algorithm. The SOC content can be accurately predicted by the laboratory spectra without spatial map. The accuracy of SOC estimation was better when based on Sentinel-2A imagery than UAV imagery, with the model determination coefficients (R 2) of 0.838 and 0.638, the root mean square error (RMSE) of 0.398 and 0.596 g kg−1, ratio of performance to deviations (RPD) of 2.485 and 1.660, and bias of − 0.028 and 0.085 g kg−1, respectively. The SOC estimation model using Sentinel-2A imagery demonstrated better stability than that using UAV imagery, with the mean uncertainty standard deviations (SD) of 0.226 and 0.291 g kg−1, respectively. The spatial SOC maps generated from both UAV and Sentinel-2A imagery presented more subtle variation information when compared with the spatial interpolation results. Although the prediction accuracy of SOC from UAV was lower than that from Sentinel-2A, UAV imagery could disclose more fine distribution information of SOC at the local level. The UAV imagery can be used to estimate SOC in coastal wetlands but with a higher cost than freely available Sentinel images. We recommend using sentinel imagery to predict SOC at regional, national, or even global levels.

The most important aspect of automatic driving and traffic surveillance is vehicle detection. In addition, poor road conditions caused by potholes are the cause of traffic accidents and vehicle damage. The proposed work uses deep learning models. The proposed method can detect vehicles and potholes using images. The faster region-based convolutional neural network (CNN) and the inception network V2 model are used to implement the model. The proposed work compares the performance, accuracy numbers, detection time, and advantages and disadvantages of the faster region-based convolution neural network (Faster R-CNN) with single shot detector (SSD) and you only look once (YOLO) algorithms. The proposed method shows good progress than the existing methods such as SSD and YOLO. The measure of performance evaluation is Accuracy. The proposed method shows an improvement of 5% once compared with the previous methods such as SSD and YOLO.

Existing segmentation and augmentation techniques on convolutional neural network (CNN) has produced remarkable progress in object detection. However, the nominal accuracy and performance might be downturned with the photometric variation of images that are directly ignored in the training process, along with the context of the individual CNN algorithm. In this paper, we investigate the effect of a photometric variation like brightness and sharpness on different CNN. We observe that random augmentation of images weakens the performance unless the augmentation combines the weak limits of photometric variation. Our approach has been justified by the experimental result obtained from the PASCAL VOC 2007 dataset, with object detection CNN algorithms such as YOLOv3 (You Only Look Once), Faster R-CNN (Region-based CNN), and SSD (Single Shot Multibox Detector). Each CNN model shows performance loss for varying sharpness and brightness, ranging between −80% to 80%. It was further shown that compared to random augmentation, the augmented dataset with weak photometric changes delivered high performance, but the photometric augmentation range differs for each model. Concurrently, we discuss some research questions that benefit the direction of the study. The results prove the importance of adaptive augmentation for individual CNN model, subjecting towards the robustness of object detection.

Unmanned Aerial Vehicle (UAV) imagery are being assessed for analyzing within field spatial variability for agricultural precision management, because unmanned aerial vehicle imagery may be acquired quickly during critical periods of rapid crop growth. The objective of this study was to evaluate the use of unmanned aerial vehicle for the monitoring barley growth. Unmanned aerial vehicle imagery obtained from middle February to late June in Wanju, Jeollabuk-do. Unmanned aerial vehicle imagery corrected geometrically and atmospherically to calculate normalized difference vegetation index (NDVI). We analyzed the relationships between NDVI UAV of barley and biophysical measurements such as plant height, number of tiller and shoot dry weight over an entire barley growth period. The similar trend between NDVI UAV and growth parameters was shown. Correlation analysis between NDVI UAV and barley growth parameters revealed that NDVI UAV was highly correlated with shoot dry weight (r=0.932) and plant height (r=0.879). According to the relationship among growth parameters and NDVI UAV , the temporal variation of NDVIUAV was significant to interpret barley growth. The spatial distribution map of barley growth was in strong agreement with the field measurements in terms of geographical variation and relative numerical values when NDVI UAV was applied to power function. From these results, NDVI UAV can be used as a new tool for monitoring barley growth.

Non‐native species maps are important tools for understanding and managing biological invasions. We demonstrate a novel approach to extend presence modeling to map fractional cover (FC) of non‐native yellow sweet clover Melilotus officinalis in the Northern Great Plains, USA. We used ensembles of MaxEnt models to map FC across landscapes from satellite imagery trained from regional aerial imagery that was trained by local unmanned aerial vehicle (UAV) imagery. Clover cover from field surveys and classified UAV imagery were nearly identical (n = 22, R2 = 0.99). Two classified UAV images provided training data to map clover presence with MaxEnt and National Agricultural Imagery Program (NAIP) aerial imagery. We binned cover predictions from NAIP imagery within each Sentinel‐2 pixel into eight cover classes to create pure (100%) and FC (20%–95%) training data and modeled each class separately using MaxEnt and Sentinel‐2 imagery. We mapped pure clover with one classification threshold and compared its performance to 15 candidate maps that included FC predictions outside pure predictions. Each FC map represented alternative combinations of five MaxEnt thresholds and three approaches to assign cover to pixels with multiple predictions from the FC ensemble. Evaluations of performance with independent datasets revealed maps including FC corresponded to field (n = 32, R2 range: 0.39–0.68) and UAV (n = 20, R2 range: 0.61–0.84) data better than pure clover maps (R2 = 0.15 and 0.31, respectively). Overall, the pure clover map predicted 3.2% cover, whereas the three best performing FC maps predicted 6.6%–8.0% cover. Including FC predictions increased accuracy and cover predictions which can improve ecological understanding of invasions. Our method allows efficient FC mapping for vegetative species discernible in UAV imagery and may be especially useful for mapping rare, irruptive or patchily distributed species with poor representation in field data, which challenges landscape‐level mapping.

In the past few years, automatic building detection in aerial images has become an emerging field in computer vision. Detecting the specific types of houses will provide information in urbanization, change detection, and urban monitoring that play increasingly important roles in modern city planning and natural hazard preparedness. In this paper, we demonstrate the effectiveness of detecting various types of houses in aerial imagery using Faster Region-based Convolutional Neural Network (Faster-RCNN). After formulating the dataset and extracting bounding-box information, pre-trained ResNet50 is used to get the feature maps. The fully convolutional Region Proposal Network (RPN) first predicts the bounds and objectness score of objects (in this case house) from the feature maps. Then, the Region of Interest (RoI) pooling layer extracts interested regions to detect objects that are present in the images. To the best of our knowledge, this is the first attempt at detecting houses using Faster R-CNN that has achieved satisfactory results. This experiment opens a new path to conduct and extent the works not only for civil and environmental domain but also other applied science disciplines.

Unmanned aerial vehicle (UAV) imagery is considered the best approach to acquire high-resolution data over a large area in a short time, such as in the case of coastal wetlands comprising small-sized patches of perennial plants. However, to the best of our knowledge, no previous studies have used UAV imagery to quantitatively estimate blue carbon in coastal wetlands. Therefore, this study aimed to estimate the area-wide carbon stocks accumulated in halophyte habitats through the case studies of the southern coastal wetlands in South Korea using UAV imagery. Accordingly, object-based image analysis was used to identify the spatial distribution of major halophytes based on the spectral reflectance derived from multispectral UAV images. Elemental analysis (EA) was performed to estimate the content of organic carbon accumulated in individual halophytes, and the halophyte communities identified in the case study area were confirmed to have accumulated 20.80 MgC of blue carbon. These results demonstrate that applying EA to UAV images can be an innovative method to solve the spatial uncertainty encountered in the monitoring of blue carbon in existing field-based surveys and traditional remote sensing.

Mangrove forests provide critical ecosystem services for coastal protection, biodiversity conservation, and carbon sequestration, yet monitoring their condition at crown level remains challenging using traditional field surveys alone. This study investigates the use of high-resolution unmanned aerial vehicle (UAV) imagery and deep learning for automated delineation of mangrove crowns in Lian, Batangas, Philippines. Orthomosaic RGB images acquired at 25 ~ m altitude (GSD ${\approx} 1.34 ~ \mathrm {cm} ^ { 2}$ pixel) were tiled into $256 \times 256$ px patches and manually annotated into three structural classes: Individual Crowns (IC), Crown Clusters (CC), and Canopy Gaps (CG). To reduce systematic confusion between gaps and low-canopy vegetation, a vegetation-focused dataset variant retaining only IC and CC labels was also constructed. We fine-tuned multiple Mask R-CNN variants (R101C4, R101DC5, R101FPN, X101-32 $\times$ 8d-FPN) pretrained on COCO and benchmarked them using Average Precision (AP) across object scales. The R101FPN backbone achieved the best trade-off between accuracy and efficiency (Scenario 1), with mean AP=17.71, $\mathrm{AP}_{50}=32.43$, and shortest training time per 1k iterations. Using this backbone, a second scenario incorporating an expanded, multi-site dataset and removal of the gap class further improved overall detection performance to $\mathrm{AP}=21.71$ and $\mathrm{AP}_{50}=46.24$, at substantially reduced training time. To quantify segmentation quality beyond box-level metrics, we computed region- and boundary-aware measures for the best model: mean Dice coefficient of 0.9273, Boundary IoU of 0.5181, and symmetric Hausdorff distance of 18.54 px over 295 validation images. These results indicate accurate recovery of crown extent, with moderate boundary imprecision in dense clusters and low-contrast conditions. Overall, the findings demonstrate that UAV-derived imagery coupled with Mask R-CNN provides a practical and effective framework for operational mangrove crown mapping in the Philippines, and highlight remaining challenges related to small crowns, crown clusters, and gap representation. -Mangrove forests, UAV imagery, Deep learning, Mask R-CNN, ResNet101, Object detection, Instance segmentation