With the development of Unmanned Aerial Vehicle (UAV) technology, it has been applied to various tasks in different fields. The prerequisite for a UAV to perform successful aerial tasks is accurate localization of its own position. Generally, traditional UAV navigation relies on the Global Navigation Satellite System (GNSS) for localization. However, this system has disadvantages such as instability and susceptibility to interference, leading to situations where UAV cannot use GNSS for positioning, known as GNSS-denied environments. This study focuses on the navigation and positioning of UAV in GNSS-denied environments and proposes a UAV visual retrieval and positioning method that comprehensively utilizes local and global deep learning features of known satellite orthophotos. Specifically, ConvNeXt is used as the backbone network, combined with generalized mean pooling, to form a retrieval feature extraction algorithm for extracting global features of satellite and UAV images. A triplet loss function considering the overlapping area between images is designed for the retrieval and positioning tasks, and a corresponding training data set is established to train the feature extraction algorithm. Then, the satellite images within a certain range are retrieved according to the extracted global features, and the preliminary retrieval results are obtained. In order to further improve the accuracy of the retrieved target images, the LoFTR algorithm based on deep learning local features is used for matching and reordering. Since the LoFTR algorithm has many mismatches, RANSAC is used to screen the matching results. Experiments using the test datasets we established demonstrate that the proposed method obtains an average accuracy of 90.9% and an average time cost of 2.22 seconds for retrieving satellite images in different seasons from fully overlapped UAV simulated images. The accuracy of the UAV real image test is 87.5%, which can meet the UAV positioning requirements.

Land surface temperature is one of the important land surface parameters that characterizes the local thermal environment. Unmanned Aerial Vehicle (UAV) thermal infrared remote sensing has the advantage of high spatial resolution, which provides data support for obtaining high-resolution local land surface temperature data. In recent years, how to accurately retrieve the surface temperature based on UAV thermal infrared remote sensing data has attracted great attention. This paper systematically explores the method of retrieving land surface temperature from UAV thermal infrared remote sensing data and synchronized atmospheric vertical profile data. We collected the UAV thermal infrared images and atmospheric vertical profile data simultaneously within the central campus of Nanjing University of Information Science and Technology and its surrounding area using the UAV-based WIRIS Pro Sc thermal imager and temperature and humidity sensor. To obtain the accurate land surface thermal radiance, the atmospheric influence on the UAV thermal infrared images was eliminated by calculating the atmospheric downward thermal radiation, upward thermal radiation, and atmospheric transmittance. Land cover data were generated from UAV multispectral data, and then the land surface emissivity was calculated based on the land cover data and emissivity spectrum library. Finally, the land surface temperature was retrieved based on the land surface thermal radiance and surface emissivity. The retrieved land surface temperature was validated by comparing with the corresponding measured land surface temperature after corrections. We also analyzed the spatial pattern of the UAV land surface temperature and the factors that affect surface temperature retrieval. The results showed that the use of synchronized temperature and humidity profiles can effectively remove atmospheric effects, ensuring accuracy of off-ground radiance measurements under varying water vapor conditions. Our retrieval method can effectively retrieve surface temperature from UAV thermal infrared images. The retrieved land surface temperature achieved a coefficient of determination of 0.91. The difference between the retrieved and observed land surface temperature ranged from 0.06 to 4.96 K, with 55.56% of the samples showing differences less than 2 K. The surface temperature showed obvious spatial variation which was closely related to the type of surface cover. Artificial surfaces such as buildings and roads had relatively high surface temperatures, generally above 325 K. Natural surfaces such as woodlands and grasslands had relatively low surface temperatures, generally not exceeding 310 K. This study provides a valuable reference for retrieving high resolution land surface temperature from UAV-based thermal infrared remote sensing data, and also provides a technological support for local thermal environment monitoring.

The use of Unmanned Aerial Vehicles (UAVs) for road image collection is advantageous owing to their large scope and cost-effectiveness. However, the size and shape of road damages vary significantly, making them challenging to predict. Furthermore, due to the limitations of computational resources, generalized target detection algorithms are only applicable to small-size images (512 pixels× 512 pixels or 640 pixels× 640 pixels). This makes them unsuitable for direct application to large-size UAV images (5 472 pixels× 3 648 pixels or 7 952 pixels × 5 304 pixels). The utilization of traditional methods for the detection of multi-scale targets in large-size images is associated with a number of issues, including the slicing of large-size targets and the failure to detect small-size targets. To address these challenges, this paper presents an innovative solution that combines the global-local multiscale fusion strategy with YOLOv5-RDD. First, a YOLOv5-RDD model is constructed, and based on the existing YOLOv5 model, a multiscale C3 (MSC3) module and a Contextual Feature Pyramid Network (CFPN) are designed to improve the detection capability of multiscale targets. Additionally, we introduce an extra detection head for larger-size targets. Then, a global-local multiscale fusion strategy is proposed, which uses resizing and slicing means to obtain global and local information of large UAV images, and then superimposes the global and local multiscale information to obtain the multi-scale information of the whole large image. The detection results are optimized using the center non-maximum value suppression algorithm. Specifically, the global-local multiscale fusion strategy first trains the YOLOv5-RDD using multiscale training strategy to learn complete multiscale features. Then, YOLOv5-RDD predicts multiscale road damages in large-size images using a multiscale prediction strategy to avoid directly applying it to these images. Finally, we use center non-maximum suppression to eliminate redundant object detection boxes. To verify the effectiveness of the proposed method and meet real-world requirements, a UAV-RDD dataset specialized for UAV road disease detection is created. The experimental results show that compared with the original YOLOv5 model, the new model YOLOv5-RDD improves the mAP by 5.8%, while the global-local multiscale fusion strategy improves the mAP by 9.73% compared with the traditional method. The MSC3 achieves the maximum enhancement of mAP@0.5, with an improvement of 2.6%, contributing only 0.8 M parameters. The CFPN yields an improvement of 0.2% in mAP@0.5 while reducing the number of parameters by 8 M. These results fully prove the effectiveness and superiority of the method in this paper.

The segmentation of fruit tree canopy based on Unmanned Aerial Vehicle (UAV) visible spectral images is greatly influenced by complex background information such as topographic relief, shrubs, and weeds. Although existing deep neural networks can improve the robustness of canopy segmentation to a certain extent, they ignore the global context and local detailed information of the canopy due to limited receptive field and information interaction, which restricts the improvement of canopy segmentation accuracy. To address these issues, this paper introduces the Canopy Height Model (CHM) and deep learning algorithms, and proposes a fruit tree canopy segmentation method that couples Convolutional Neural Networks (CNN) and Attention Mechanisms (AM) based on UAV photogrammetry. This method first constructs a coupled deep neural network based on CNN and AM through transfer learning to extract both the local and global high-level contextual features of fruit tree canopies. Meanwhile, considering the correlation between deep semantic features and the position information of fruit tree canopies, a local and global feature fusion module is designed to achieve collaborative tree canopy segmentation of attributes and spatial positions. Taking the citrus tree canopy segmentation as an example, the experimental results demonstrate that the use of the CHM can effectively suppress the influence of topographic relief. Our proposed method can also significantly reduce the interference of underlying weeds or shrubs on canopy segmentation, and achieves the highest Overall Accuracy (OA), F1 score, and mean Intersection over Union (mIoU) of 97.57%, 95.49%, and 94.05%, respectively. Compared with other state-of-the-art networks such as SegFormer, SETR_PUP,TransUNet, TransFuse, and CCTNet, the mIoU obtained by the proposed method increases by 1.79%, 8.83%, 1.16%, 1.43%, and 1.85%, respectively. The proposed method can achieve high-precision segmentation of fruit tree canopies with complex background information, which has important practical value for understanding the growth status of fruit trees and fine management of orchards.

Soil Organic Matter (SOM) content is an important indicator for measuring soil fertility and has a significant impact on food production. China's coastal saline-alkali lands are vast, and SOM exhibits significant spatial heterogeneity. The traditional method of “field collection and experimental measurement” is time-consuming and labor-intensive, making it difficult to quickly depict the spatial distribution characteristics of SOM content in saline-alkali areas. Therefore, efficiently and accurately using remote sensing technology to invert SOM content can provide insights into soil fertility, so as to help adjust and optimize agricultural production and management. This study focused on the bare soil (0~10 cm soil layer) in a typical coastal saline-alkali land of the Yellow River Delta and constructed remote sensing indices from the Unmanned Aerial Vehicle (UAV) remote sensing spectral and spatial texture information. The multi-linear stepwise regression model, partial least squares model, and random forest model were employed to quantitatively estimate SOM. We compared and analyzed the impact of cultivated and uncultivated soil on SOM content retrieval through UAV remote sensing methods, and also explored whether the integration of remote sensing information sensitive to soil salt content aids in SOM retrieval in saline-alkali areas. The results show that: (1) In the inversion of SOM content, the accuracy of random forest model (R² ranging from 0.83 to 0.95) was significantly higher than that of multiple linear stepwise regression (R² ranging from 0.26 to 0.69) and partial least squares model (R² ranging from 0.37 to 0.72); (2) Compared to uncultivated soil (R² ranging from 0.26 to 0.95), the inversion accuracy of organic matter in cultivated soil (R² ranging from 0.54 to 0.94) was significantly higher. That is to say, cultivation treatment enhanced the response of spectral indices to SOM content, thereby increasing the inversion accuracy of SOM content, which provides new insights for precise monitoring of SOM; (3) Integrating remote sensing texture information or soil salt content information (i.e., salt content and salt sensitivity) can significantly increase the inversion accuracy of SOM. This study provides theoretical and technical support for improving SOM content inversion in salinized coastal farmland at the field scale, ultimately contributing to the development of modern agriculture in salinized coastal areas.

At present, the classification of urban vegetation is affected by similar characteristics and similar spectra, resulting in misclassification of vegetation. Therefore, a Multi-scale Feature Perception Network (MFDN) combined with high-resolution UAV visible light images is designed to classify urban vegetation. This network addresses the problem of misclassification and reduces the loss of spatial information by introducing coordinate convolution in the input layer. It constructs parallel networks to enhance multi-scale feature information and introduces repeated multi-scale fusion modules between networks to maintain high-resolution representation in the entire process and reduce the loss of detailed features. In addition, the separation feature module is added to expand the receptive field and obtain multi-scale features, thereby effectively alleviating the phenomenon of misclassification and omission of urban vegetation. The results show that the MFDN method improves the classification of urban vegetation mainly through spatial patterns rather than spectral information from UAV visible light images. The average overall accuracy is 89.54%, the average F1 score is 75.85%, and the average IOU is 65.45%. The segmentation results are accurate and complete. Therefore, the proposed method is compatible with the easy-to-operate low-cost UAV system, is suitable for rapid survey of urban vegetation, and can provide technical support and scientific basis for urban space utilization and ecological resource survey.

Accurate identification of ground fissures in mining areas is significant for disaster prevention, mitigation, and ecological environment restoration. In this study, a ground fissure extraction method is proposed based on the improved active contour model for UAV images in mining areas, aiming at accurately extracting ground fissures from high-resolution UAV images. Firstly, the Otsu algorithm was used to calculate the background and initial values of ground fissures as prior knowledge. Secondly, the extraction energy functions of the background and initial values of ground fissures were constructed and introduced into the traditional CV active contour model to enhance the pertinence of ground fissures extraction. Finally, ground fissures were extracted through the continuous evolution of the contour. Based on UAV images obtained in Dalai Nurg mining area, Inner Mongolia, the improved active contour model was used to extract ground fractures, and compared with traditional Canny edge detection algorithm, Support Vector Machine (SVM), Maximum Likelihood Method (MLM), and traditional CV active contour model methods for analysis and accuracy evaluation. The results show that the traditional Canny edge detection algorithm and traditional CV active contour model had the poor extraction in a small area with a single type of land cover. The improved active contour model, SVM, and MLM had achieved good results, and the improved active contour model method had the highest accuracy. In addition, in a large area with relatively complex land cover types, the traditional methods such as Canny edge detection algorithm, SVM, MLM, and CV active contour model had many omissions and errors, and the kappa coefficient was lower than 0.7. However, the improved active contour method still achieved better results, and the Kappa coefficient was about 0.9. Therefore, the proposed method could effectively improve the accuracy and stability of ground fissure extraction by introducing prior knowledge.

In recent years, the conflict between rapid growth of economy and protection of water resources has become increasingly prominent. Water bloom has also become an important environmental issue, both domestically and internationally. Efficient use of remote sensing to predict water bloom outbreaks is of great significance to promote the management and protection of the lake and reservoir water environment. Using multispectral images of Unmanned Aerial Vehicle and measured water quality parameters as data sources, the water quality parameters were inversed by the Matching Pixel-by-pixel (MPP) algorithm, and the water bloom information was extracted through the Normalized Difference Vegetation Index (NDVI) threshold method. Then, a short-term prediction model of water bloom based on geographical weighted regression was proposed. The characteristic of this model was that it can accurately estimate the area and spatial location of the water bloom within a short period. We also discussed the influence of window size on the prediction results. The results show that: (1) based on the proposed model, the prediction accuracy of bloom area reached 96.19%. For the spatial distribution of water bloom, the overall classification accuracy of water bloom and non-water bloom pixels was both greater than 0.969, and the producer accuracy and Kappa coefficient were both higher than 0.5; (2) the total nitrogen, total phosphorus, and dissolved oxygen concentrations inverted using the MPP algorithm showed high correlation with the measurements (R² was 0.888 6, 0.854 6, and 0.896 9, respectively). The water bloom information extracted by NDVI threshold method had a high consistency with the true color orthophoto-based visual results in terms of bloom outbreak density and spatial distribution; (3) the prediction window size was closely related to the spatial resolution of data. The prediction window of different sizes directly affected the accuracy of water bloom prediction results. Compared with the window size of 8×8, 12×12, and 14×14, the highest accuracy (0.98), precision (0.77), and Kappa (0.77) were obtained with a window size of 10×10. Hence, the 10×10 prediction window was the most suitable for the water bloom prediction in this study area. The model developed here can effectively predict water bloom in lakes and reservoirs in advance, which provides a reference for improving short-term water bloom prediction and warning.

Each year, approximately 2.1 million sheep are released to graze freely in vast, forest-covered, and mountainous areas throughout Norway. At the end of the grazing season, farmers must find and round up their sheep. This can be a time consuming and challenging process because of the large area and cluttered nature of the sheep grazing environment. Existing technologies that help farmers find their sheep, such as bells, radio bells, electronic ear tags and UAVs, are limited by the cost, signal coverage, and low degree of automation, which cannot efficiently and automatically locate sheep in the wild. This study proposes an automatic sheep detection algorithm using UAV images. A model architecture using the ResNet and ResNeXt as the backbone networks is designed to address the automatic sheep detection task from UAV RGB and infrared images. Our study evaluates how well this model meets performance and processing speed requirements of a real-world application. We also compare models using fused RGB and infrared data to models using either RGB or infrared as input, and further explore the model complexity and generalization ability. Results show that fusion of RGB and infrared data yields better average precision results than using single RGB or infrared dataset in the model. The set of optimal solutions achieve average precision scores in the range of 69.6% to 96.3% with inference times ranging from 0.1 to 0.6 seconds per image. The most accurate network achieves a grid precision of 97.9% and a recall of 90.1%, at a confidence threshold of 0.5. This corresponds to the detection of 97.5% of the sheep in the validation dataset. These satisfactory results demonstrate the great potential of the proposed automatic sheep detection method using multi-channel UAV images for improving sheep roundup.

Wetland plant species diversity, as a quantifiable indicator reflecting the level of organization in an ecosystem’s community, can reveal the community organization and stability of wetland ecosystems. Accurate assessments of wetland health, degradation, and restoration status are crucial for effective wetland management and protection. Therefore, timely understanding of the current status of wetland plant community species diversity is of great importance. However, traditional field survey methods are time-consuming and labor-intensive, limited by temporal costs, and cannot achieve large-scale synchronous observation. Meanwhile, hyperspectral technology, with its high resolution, can capture more abundant spectral information, providing an opportunity for the realization of this goal. To investigate how to accurately invert wetland plant species diversity using hyperspectral technology, we investigated the wetland plants in Hanzhong Crested Ibis National Nature Reserve in Shaanxi Province and simultaneously acquired hyperspectral images of the plant canopy. Species diversity was characterized by four indicators: Simpson (D_S), Margalef (D_M), Shannon-Weiner (H'), and Pielou (J). The inverse model was established using three methods: Random Forest (RF), Back Propagation Neural Network (BPNN), and Partial Least Squares (PLS). Finally, the inverse projection of regional species diversity was realized. The outcomes indicate that spectral differentiation complicates the association between spectra and species diversity indices, producing a range of sensitive bands. Notably, the first-order differential transform is superior in extracting sensitive bands compared to the second-order differential transform. Furthermore, correlating species diversity indices can be enhanced through the integration of vegetation indices from various bands. When applying the RF model to analyze differential spectra and vegetation indices, it was found that both using original features and combinations of features, the model's inversion results demonstrated similar and high accuracy (R² > 0.40). Particularly, in predicting H' and J, the model exhibited strong precision (R² > 0.6), and in terms of D_S, R² also exceeded 0.5, indicating potential predictive capabilities. However, in reverting another measurement of D_M, the model showed lower accuracy (R² < 0.5), suggesting challenges in improving the model's predictive power. This study demonstrates the effectiveness of UAV hyperspectral technology in the accurate inversion of wetland plant species diversity and confirms the reliability of the method for species diversity inversion at the Unmanned Aerial Vehicle (UAV) scale, achieved through spectral differential transformation and feature variable extraction combined with the random forest model. This technique can provide technical support for the large-scale detection of wetland biodiversity and offer references for decision-making by relevant management departments.

[Objectives] Cross-view image geolocation refers to a technology that determines the geographical location of an image by matching it with reference images taken from different perspectives and possessing precise location information. This technology plays a crucial role in real-world applications such as Unmanned Aerial Vehicle (UAV) navigation, environmental monitoring, and target positioning. Currently, most deep learning-based cross-view image retrieval and geolocation methods for drone-satellite tasks rely heavily on supervised learning. However, the scarcity of high-quality labeled data presents a significant limitation, hindering the generalization capability of these models. Moreover, existing methods often fail to effectively model the spatial layout of images, making it difficult to bridge the substantial domain gap between cross-view images, thereby limiting the accuracy and robustness of geolocation tasks. [Methods] To address these challenges, this paper proposes a novel cross-view image retrieval and localization architecture called DINO-MSRA. The architecture first employs the DINOv2 large model framework, fine-tuned by Conv-LoRA, as the feature encoder. This enhances the model's feature extraction capabilities with fewer parameters, improving both efficiency and accuracy. Second, we design a spatial relation-aware feature aggregator based on the Mamba module (MSRA) to more effectively aggregate image features. By embedding spatial configuration features into the global descriptor, this module significantly improves the model's performance in cross-view matching tasks, especially in complex scenarios where spatial relationships between objects are crucial. Finally, the InfoNCE loss function is adopted to train the model, optimizing contrastive learning and ensuring more accurate retrieval and localization results. [Results] Extensive comparative and ablation experiments were conducted on the University-1652 and SUES-200 datasets. The experimental results show that for drone-view target localization (drone→satellite) and drone navigation (satellite→drone) tasks, the proposed method achieves R@1 accuracies of 95.14% and 97.29%, respectively, on the University-1652 dataset, representing improvements of 0.68% and 1.14% over the current best algorithm, CAMP. On the SUES-200 dataset at an altitude of 150 meters, R@1 accuracies reach 97.2% and 98.75%, which are 1.8% and 2.5% higher than CAMP, respectively. Moreover, the proposed method requires significantly fewer parameters than existing algorithms, only 19.2% of those used by Sample4Geo. [Conclusions] In summary, the proposed DINO-MSRA architecture outperforms current state-of-the-art methods in cross-view image matching, achieving higher accuracy and faster inference speed. These results demonstrate its robustness and practical application potential in challenging real-world scenarios.

[Objectives] Unmanned Aerial Vehicle (UAV) and satellite remote sensing technologies have been successfully applied to estimate soil organic carbon and other attributes. However, their application to soil texture estimation remains relatively limited, highlighting the need for further research in this area. This study focuses on three farmland plots located in Zhuozhatan Village (Huzhu County), Nilongkou Village (Lalongkou Town, Huangzhong District), and Baitu Village (Lushar Town, Huangzhong District) within the Huangshui River Basin of Qinghai Province. It explores the potential of UAV and satellite remote sensing technologies for estimating soil texture content at the field scale. [Methods] Using UAV platforms equipped with two hyperspectral cameras, field-scale imaging of farmland soils was conducted. Additionally, a field spectrometer was used to collect in-situ soil spectra, and a total of 838 soil samples were collected from 2022 to 2024. Satellite imagery was also obtained for the same time periods, including GF1/2/7 (Gaofen 1/2/7), Sentinel-2A, and ZY1-02D (Ziyuan 1-02D). Laboratory analyses determined soil particle size distribution and acquired indoor soil spectral data. Based on these datasets, statistical modeling and soil texture content estimation were performed using the XGBoost (Extreme Gradient Boosting) method for laboratory, field in-situ, UAV, GF, ZY1-02D, and Sentinel-2 spectral data. Spatial distribution maps of soil texture content were then generated. [Results] ① Among the XGBoost model results, the highest model accuracy for UAV image spectra achieved an RPD (Ratio of Performance to Deviation) of 2.441, while the optimal RPD values for GF1/2/7, ZY1-02D, and Sentinel-2 satellite imagery were 1.815, 1.601, and 1.561, respectively. ② The estimation accuracy based on UAV and satellite imagery was lower than that derived from field spectrometer measurements. The accuracy ranking was as follows: laboratory spectra > field in-situ spectra > UAV image spectra > GF1/2/7 satellite image spectra > ZY1-02D satellite image spectra > Sentinel-2 satellite image spectra. Among soil texture components, clay content estimation showed the highest accuracy (RPD = 2.70), followed by silt (RPD = 2.24) and sand (RPD = 1.91). ③ Sand and clay content exhibited a negative correlation with soil spectral reflectance, whereas silt content displayed a positive correlation. The sensitive bands for sand, silt, and clay content were primarily concentrated in the near-infrared region (780~2 400 nm). ④ The content of sand, silt, and clay exhibited minor variations over three years, demonstrating relative stability. The mapping results for the three plots showed soil texture contents predominantly in the following ranges: 67% < sand ≤ 83%, 10.6% < silt ≤ 19.1%, and 3.2% < clay ≤ 6.6%. [Conclusions] At the field scale, UAV imagery was identified as the most effective data source for soil texture content mapping, providing strong support for precision agricultural management. While GF1/2/7 and ZY1-02D satellite imagery were found to be sufficient for texture mapping, Sentinel-2 satellite imagery was too coarse for field-scale mapping.

[Objectives] Multiple Unmanned Aerial Vehicles (UAVs) path planning is a pivotal technology enabling efficient and cooperative operation of UAV swarms in complex environments. Fundamentally, it constitutes a constrained multi-objective optimization problem characterized by a sparse feasible region. Due to their robust global search capabilities, evolutionary algorithms have been widely adopted to address this class of problems. However, existing methodologies frequently neglect the spatial structural attributes of paths during the initialization phase and primarily rely on generic stochastic search operators for path regeneration. These limitations restrict their ability to effectively exploit the fitness function and the spatial relationships between UAVs and their operational environment, thereby hindering the generation of high-quality, globally feasible solutions under constrained computational resources. [Methods] To address these challenges, this paper introduces a Multi-Source Heuristic Evolutionary Algorithm (MSHEA) for multiple UAV path planning. MSHEA systematically integrates multi-source heuristic information related to spatial path structure, fitness data, and environmental context to enhance both the path initialization and regeneration processes. Specifically, we propose a Sequential Directed Expansion-based Initialization (SDEI) strategy to generate high-quality initial paths that exhibit spatial rationality and structural diversity. Furthermore, we develop a path regeneration mechanism integrating fitness and Flight Environment Information (FFEI), which substantially improves the repair efficiency of infeasible paths and the local optimization of feasible ones. [Results] Empirical validation was conducted using eight publicly available benchmark datasets for multiple UAV path planning, covering a range of complexity levels. The experimental results demonstrate MSHEA's superior performance and elevated stability across diverse flight scenarios: (1) Compared to the suboptimal benchmark algorithm, MSHEA achieved a 1%~6% improvement in hypervolume and a 6%~81% reduction in inverted generational distance. (2) Both the SDEI and FFEI components were confirmed to have significant positive impacts on the algorithm's overall performance. (3) MSHEA shows low sensitivity to its newly introduced hyperparameters, indicating strong adaptability and generalization capability. [Conclusions] In conclusion, MSHEA improves the effectiveness of solving multiple UAV path planning problems by incorporating multi-source heuristic information. This leads to robust and reliable performance for the challenges inherent in collaborative UAV missions.