Next Article in Journal
Mission Design and Validation of a Fixed-Wing Unmanned Aerial Vehicle for Environmental Monitoring
Previous Article in Journal
Use of Simulation for Pre-Training of Drone Pilots
Previous Article in Special Issue
Visual Servoing for Aerial Vegetation Sampling Systems
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Drones
Volume 8
Issue 11
10.3390/drones8110639
drones-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Mapping Invasive Species Pedicularis and Background Grassland Using UAV and Machine Learning Algorithms

by
Jin Zhao
1,2,*,
Kaihui Li
2,3,
Jiarong Zhang
2,4,
Yanyan Liu
2,3 and
Xuan Li
5
1
Xinjiang Key Laboratory of RS&GIS Application, Urumqi 830011, China
2
Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi 830011, China
3
Bayanbulak Alpine Grassland Observation and Research Station of Xinjiang, Urumqi 830011, China
4
University of Chinese Academy of Sciences, Beijing 100049, China
5
Center for Grassland Biological Disaster Prevention and Control of Xinjiang Uygur Autonomous Region, Urumqi 830000, China
*
Author to whom correspondence should be addressed.
Drones 2024, 8(11), 639; https://doi.org/10.3390/drones8110639
Submission received: 30 July 2024 / Revised: 11 October 2024 / Accepted: 22 October 2024 / Published: 4 November 2024
(This article belongs to the Special Issue Application of Uncrewed Aerial Vehicles (UAVs) in Vegetation Monitoring)
Browse Figures
Versions Notes

Abstract

:
The rapid spread of invasive plants presents significant challenges for the management of grasslands. Uncrewed aerial vehicles (UAVs) offer a promising solution for fast and efficient monitoring, although the optimal methodologies require further refinement. The objective of this research was to establish a rapid, repeatable, and cost-effective computer-assisted method for extracting Pedicularis kansuensis (P. kansuensis), an invasive plant species. To achieve this goal, an investigation was conducted into how different backgrounds (swamp meadow, alpine steppe, land cover) impact the detection of plant invaders in the Bayanbuluk grassland in Xinjiang using Random Forest (RF), Support Vector Machine (SVM) and eXtreme Gradient Boosting (XGBoost) with three feature combinations: spectral band, vegetation index (VI), and spectral band + VI. The results indicate that all three feature combinations achieved an overall accuracy ranging from 0.77 to 0.95. Among the three models, XGBoost demonstrates the highest accuracy, followed by Random Forest (RF), while Support Vector Machine (SVM) exhibits the lowest accuracy. The most significant feature bands for the three field plots, as well as the invasive species and land cover, were concentrated at 750 nm, 550 nm, and 660 nm. It was found that the green band proved to be the most influential for improving invasive plant extraction while the red edge 750 nm band ranked highest for overall classification accuracy among these feature combinations. The results demonstrate that P. kansuensis is highly distinguishable from co-occurring native grass species, with accuracies ranging from 0.9 to 1, except for SVM with six spectral bands, indicating high spectral variability between its flowers and those of co-occurring native background species.

1. Introduction

Grasslands play a vital role in supporting the livestock economy and maintaining biosphere stability. They harbor substantial biodiversity and provide invaluable ecosystem services that are in high demand by society. Proper management can offer essential insights for regional husbandry production and land utilization [1]. Nevertheless, grassland primary productivity and habitats face significant threats from invasive species [2,3]. Over the last few decades, invasive plants have encroached upon the natural grasslands of Western China [4], causing substantial harm to the local environment and economy [5,6,7,8].
Monitoring and mapping invasive species in grasslands are imperative [9]. The distinctive spectral characteristics of invasive species compared to coexisting native plant species necessitate thorough investigation, as they frequently contribute to diminished biodiversity and the extinction of native species. Well-structured classification datasets of grasslands, with high spatial and temporal resolutions, are indispensable for scientific research [10].
P. kansuensis, a semi-parasitic plant, is predominantly distributed in Xinjiang, southwestern Gansu, Qinghai, and western Sichuan [11]. This species exhibits strong invasive potential and can rapidly expand within grassland ecosystems [12]. As a consequence of its spread, there has been reduced utilization of grasslands and increased competition with other pasture plants for water and nutrients, resulting in decreased yields of high-quality pasture plants and severe impacts on local animal husbandry development [13]. Wangdan [11] predicted that climate change will further influence the distribution of P. kansuensis habitat by the 2050s and 2070s, with continued expansion in western China. Preserving these grasslands and implementing effective measures for their restoration are essential [14]. Understanding the spatial and temporal distribution of invasive plant species within specific habitats is essential for comprehending invasion patterns [15]. Therefore, identifying the distribution and spectral characteristics of P. kansuensis is crucial for its prevention and control.
Due to limitations in spatial and spectral resolution, differentiating between various species of grassland using satellite remote sensing images remains challenging [16,17,18]. Furthermore, in the context of grasslands, the identification and recognition of these species can be challenging because of their spectral properties resembling those of native species or growing alongside them, making it hard to visually distinguish them from the surrounding “background”.
In recent years, uncrewed aerial vehicle (UAV) technology has become increasingly popular in the remote sensing community, which is attributable to its ability to consistently capture the differentiation of spectral characteristics among different species and significantly improve the accuracy of identifying species within smaller vegetation units [19] with very high spatial resolution [20]. In contrast to satellite or airplane-based remote sensing, UAVs operate at low flight altitudes, which reduces weather-related limitations, lowers operational costs and complexity, and allows for flexible deployment for repeat missions. As a result, this technological approach provides valuable support for effectively monitoring grassland vegetation species composition [21]. Literature analysis indicates that after 2010, particularly after 2015, there has been an exponential growth in UAV remote sensing and related research on grasslands [22,23,24,25,26]. Several researchers have utilized UAVs equipped with hyperspectral imagery, field surveys, and high-resolution satellite imagery to assess the species composition of grassland vegetation [27,28,29]. The effective application of remote sensing methods requires an understanding of the spectral differences between invasive alien plants and native species [30,31]. Previous research has demonstrated that the effective differentiation of subtle spectral variations among species is best achieved through the use of hyperspectral sensors, which offer numerous contiguous narrow wavelengths [32,33,34]. However, the complex nature of hyperspectral data, with its high dimensionality, poses challenges for calibration and entails substantial computational expense. Moreover, there is a scarcity of spaceborne hyperspectral sensors, and conducting commercial airborne campaigns to collect such data proves to be expensive. Additionally, efforts to upscale field-scale in situ hyperspectral data for exploration at the satellite image level have been limited. As a result, there is a lack of operational flexibility in handling hyperspectral data, which precludes large-scale mapping of invasive plants. To overcome these limitations associated with using hyperspectral datasets, researchers are investigating low-cost or freely available satellite and airborne imagery [15,35,36]. Satellite and airborne image classification have shown excellent results in mapping invasive woody [37] and shrubby [38] species. Integrating multiple features can enhance classification accuracy. However, selecting too many features may impact classification efficiency [39]. Research [39] suggests that using a feature selection method to filter the optimal combination can improve classification efficiency and accuracy. Wang’s study introduces an exceptionally effective and accurate algorithm for extracting Pedicularis from UAV imagery [40]. Nevertheless, constrained by the substantial expense associated with acquiring UAV data, extraction trials were only confined to a specific site [40]. Subsequent mapping endeavors ought to encompass varied species contexts, given the auspicious findings from preliminary assessments. It is imperative to scrutinize the capacity of UAV-derived imagery in assessing floral coverage of P. kansuensis across complex vegetative environments and diverse hues exhibited by flowering plants.
The major objectives of this study were the following: (1) to assess the relative significance of bands and indices, followed by the identification of influential bands and vegetation indices for discriminating between P. kansuensis flowers and grassland land cover across diverse grassland species using UAV imagery; and (2) to compare three machine learning models and ascertain the optimal model.

2. Materials and Methods

2.1. Study Area and Framework

The Bayanbulak area is located deep within the Central Tienshan Mountains (Figure 1), spanning from east longitude 82°27′–86°17′ to north latitude 42°18′–43°34′. As its name suggests, the region is rich in springs: in Mongolian, Bayanbulak translates to “many springs.” It covers an approximate length of 270 km and a width of 136 km, consisting of two intramontane basins, namely the Qong Yulduz and the Kigik Yulduz Basins, with a total area of 2,383,500 hectares. The basin bottoms are situated at elevations ranging from 2390–2500 m, while the surrounding high mountains reach elevations ranging between 4000–5000 m. The annual average air temperature is −4.7 °C, and the annual mean precipitation at the basin bottoms is recorded at 276.2 mm. Administratively, this region falls under Hejing County in the Bayingolin Mongol Autonomous Prefecture of Xinjiang, China. With vast expanses of alpine steppe and meadow serving as pastureland, this study region represents Xinjiang’s most extensive stock-raising base and even one of China’s most expansive and productive pasturelands. Since the 1970s, approximately 1.5 million sheep units have been grazing here annually.
For this study, a UAV M300 RTK (DJI Technology Co., Ltd., Shenzhen, China) equipped with a 6-band multispectral digital camera was used to assess species composition and presence of P. kansuensis in the Bayanbulak grassland located in Xinjiang province, China (Figure 1). To ensure consistency across diverse locations, a standardized processing workflow was established for handling UAV-acquired imagery (Figure 2). In the field data, RTK localization and species identification on the ground, along with quadrangle monitoring, provide essential prior knowledge for localization and classification, as well as training data for UAV data acquisition and information extraction (Figure 2). The corrected images were then employed for machine learning-based species classification analysis. Vegetation indices mainly encompassed normalized difference vegetation index and ratio vegetation index for leaf and canopy characteristics (Table 1). Subsequently generated maps illustrating species distribution were further utilized to examine the spatial spread of P. kansuensis.

2.2. Field Data

This study categorized three UAV-captured images acquired at three plots of 100 × 100 m during the flowering period to investigate P. kansuensis. The main communities and species were identified. The diverse land cover and grass species led to varying spectral backgrounds in the alpine steppe and swamp meadow (Table 2). All three plots were at their peak flowering phase. The distinction between the two alpine steppes lay in the fact that one was situated adjacent to a river with water, so the grassland exhibited a high level of green coverage, and the vegetation was in excellent condition, whereas the other was positioned near an arid riverbed with no water, thus was characterized by green vegetation and litter. Three 1 × 1 m quadrats on the ground were utilized to assess distortion in the orthoimage and to identify various species (Figure 3). In each 1 × 1 m quadrat, a 50 × 50 cm quadrat was selected for vertical photography. The inflorescences of P. kansuensis were extracted, and the number of inflorescences was enumerated.

2.3. UAV Orthoimages

The DJI M300 RTK (DJI Technology Co., Ltd., Shenzhen, China) serves as UAV flight platform in this study. The DJI M300, equipped with an MS600 Pro multispectral camera (Yusense, Inc., Qingdao, China) was employed to capture 6-band images on 2 and 3 August 2022. Three flight survey missions were conducted during the flowering season to analyze spectral variations across the landscape and examine land cover composition patterns as well as potential species encroachment. Depending on the local terrain, the UAV operated at a flying height of 13.9 m, resulting in a ground resolution of 1 cm. Each image had a single-band resolution of 1280 × 960. The flight mission was planned using the DJI Pilot app, and the camera automatically captured nadir-looking images with a configuration for 75% image overlap both forward and sideways. The flight sessions occurred between 10:00 and 14:00 under clear weather conditions. The entire system had an approximate weight of 5 kg. Two intelligent flight batteries, each with a capacity of 5935 mAh, were used, resulting in a flight duration of approximately 20 to 25 min.
The camera band parameters for the MS600 Pro are outlined in Table 1. Throughout the UAV flights, a D-RTK2 high-precision GNSS field mobile station (DJI Technology Co., Ltd., Shenzhen, China) was used, which provided centimeter-level accuracy [41], thereby enhancing direct georeferencing precision and reducing the need for acquiring several GCPs on the ground. Multispectral reflectance correction and band layer stacking was conducted with YusenseRef. The effect of light on multispectral images was reduced through standard gray plate calibration and downlink light sensor (DLS) correction, ensuring the authenticity of feature spectral reflectance. Before and after each flight, images of the calibration plate were taken and were used in the radiation calibration processing of multispectral data in YusenseRef. It was placed horizontally and directly facing the sunlight to avoid being obscured while taking images. During the flight, the DLS mounted horizontally on top of the UAV synchronously measured the ambient light corresponding to the six bands and recorded it in the metadata of the captured images, which was used to automatically calibrate the light changes. The Pix4D mapper V 4.5 (Pix4D SA, Lausanne, Switzerland) was used to perform orthorectification and DSM generation. The workflow of 3D reconstruction and product generation included feature recognition, matching, triangulation (pose estimation), sparse point cloud generation, point cloud densification, 3D modeling, orthorectification and DSM generation. The final sizes of the orthorectified images were as follows: swamp meadow (8 GB), alpine steppe1 (2 GB), and alpine steppe2 (5 GB).

2.4. Machine Learning Algorithms

The machine learning algorithms RF, SVM and XGBoost were applied for classification analysis in this study. These classifiers are highly effective ensemble machine learning algorithms [19,42,43]. The sample data were partitioned into a 7:3 ratio for classification and validation purposes. Classification models were established using Python 3.9.6. In this study, the RF settings encompassed a maximum tree depth of 10 with a maximum number of trees set at 500. In the SVM classifier, a radial basis function, which is commonly used for classifying remotely sensed data, was used to map the training samples to a high-dimensional space according to the SVIs’ information and find the optimal hyperplanes. Two parameters were optimized: C = 0.1 and gamma (γ) = 4. XGBoost has demonstrated its superiority over RF in previous research [42,43]. Number of trees and other parameters varied with three plots and feature groups in XGBoost. The number of training samples was 5727 for swamp meadow, 2064 for alpine steppe1, and 3603 for alpine steppe2.

2.5. Feature Importance and Accuracy Evaluation

In this study, impurity was employed to assess feature importance in the RF classifier. Feature importance denotes the extent of relevance of a feature to the target variable and can be calculated using various methods. It is defined as the overall reduction in impurity across all nodes averaged over all ensemble trees. The primary advantage of this approach lies in its ability to compute feature importance during random forest training. The sum of permutation importance for all features would equal 1 to facilitate comparison. Five hundred repetitions were performed to obtain average values, which serve as the ultimate scores for feature importance.
The optimal classification model was chosen based on their overall accuracy, precision (or user accuracy), and recall (or producer accuracy) [43]. Global assessment of classification was performed using overall accuracy. Precision denotes the probability of an object classified in a specific class actually belonging to that class and reflects commission errors. Recall signifies sensitivity by indicating the likelihood of an object being excluded from the class, and represents omission errors.

3. Results

3.1. Classification Results and Accuracy Evaluation

Classification result maps of three plots are shown in Figure 4. Three classification models achieved acceptable accuracy (Figure 5). The overall accuracy (OA) of the three classification maps ranges from 0.77 to 0.94. XGBoost performed best, with a mean accuracy of 0.94, followed by RF with a mean accuracy of 0.93. SVM performed the worst. SVM exhibits relatively limited capabilities in identifying P. kansuensis and other flowering plants with 6-spectral bands.
When examining individual field plots, alpine steppe1 and alpine steppe2 exhibited the highest average accuracy, followed by swamp meadow. Comparing OA of land cover maps for different tested feature groups (spectral bands, VI, spectral band + VI combinations) within the same UAV flight reveals no significant differences in OA except for 6-spectral bands with SVM. The OA of spectral bands is close to that of VI and spectral band + VI. In terms of land cover classification accuracies for individual land covers, precision ranges from 0.68 (white flowering) to 1 (grass, P. kansuensis, water), while recall ranges from 0.65 (purple flowering) to 1 (grass, P. kansuensis, water) in RF classifier.
The results indicate that P. kansuensis can be distinguished with high accuracy from co-occurring native grass species, with a range of accuracy from 0.9 to 1 (Figure 5) except by the SVM model, demonstrating high spectral variability of the flower compared to co-occurring native species. Regarding input layers, accuracies are generally highest for spectral bands followed by VI and spectral + VI combinations. In terms of accuracy, commission and omission errors are generally smallest for water followed by P. kansuensis and green grasses; however, when considering soil including saline content, both precision and recall are low in swamp meadow. Evaluation of purple flowering showed that precision was generally higher than recall, indicating fewer errors in commission than omission, whereas white flowering exhibited a contrary trend.

3.2. Feature Importance

Figure 6 presents a comprehensive overview of the feature importance of the RF model, encompassing all six spectral bands, characteristic indices, and combined bands + indices. It is observed that nearly 25 bands exhibit significance in terms of feature importance. The distribution of feature importance scores varies with changes in background across three plots.
The ranking of band models based on feature importance is as follows: Red Edge 750 nm > Green > Red > NIR > Blue > Red Edge 720 nm. Among the six bands, Band 5 (red edge 750 nm) ranks highest in terms of importance score at two plots, while Band 2 (green band) and Band 3 (red band) rank within the top three across all three plots. This suggests that the red edge 750 nm, green, and red bands may be highly sensitive to species and invasive plants. Additionally, the NIR band demonstrates high importance scores in feature importance calculations for alpine steppe1. Conversely, blue band and red edge (720 nm) band exhibit relatively lower levels of importance. Notably, an inverse relationship exists between the importance of red edge 720 nm and 750 nm, with red edge 750 nm ranking highest while red edge 720 nm ranked lowest. The top index rank varies according to plant type and land background. B24 SR-RG and B10-NDRG, based on red and green bands, respectively, demonstrate high feature importance scores in swamp meadow and alpine steppe1. In contrast, B16-S (based on blue and green bands) and B5 (red edge 750 nm) show high feature scores in swamp meadow as well as alpine steppe2. Furthermore, B9 NDVI and B18 GRVI are significant in alpine steppe1 due to the presence of litter and green grass, which can be effectively separated using red or NIR bands. Characteristic indices such as Band22 Ci750 and Band23 EVI display low feature importance scores across all three plots, indicating no apparent differences in species spectrum features among UAV images.

4. Discussion

4.1. Suitability of Machine Learning Methods, UAV and Phenological Period

The suitability of machine learning methods, UAVs and phenological periods has been widely recognized. RF and XGBoost are popular choices for classification tasks due to their robust generalization performance and lower risk of overfitting, as demonstrated by numerous studies. Although the accuracy of RF is close to XGBoost, XGBoost’s parameters are more complex and require more time to process. Research has shown that the RF classifier outperformed SVM and logistic regression in classifying species occurrence [44,45]. This indicates that methods combining UAV imagery with machine learning have the potential to become valuable tools for large-scale, standardized, and cost-effective monitoring of flower cover of P. kansuensis. The core idea of SVM is to map linearly inseparable data in the original space to high-dimensional feature space through a kernel function so that the linear analysis can be carried out in the high-dimensional space. Therefore, 6-band low-dimensional features showed relatively low accuracy.
The study confirmed that utilizing UAV images along with an RF and XGBoost machine learning algorithm could effectively detect P. kansuensis and achieve excellent results using only 6 spectral bands. Furthermore, UAV-based methods hold great promise for grassland monitoring because of their affordability [46], enabling various operators (e.g., researchers, farmers, ecologists) to obtain high spatial resolution data from different sensors simultaneously while covering large areas within a limited time frame. Over time, they could be employed with an “on-demand” approach to capture specific stages of plant growth patterns (such as flowering time), particularly in areas with substantial cloud cover [15,23]. Nevertheless, several challenges need to be addressed before the widespread implementation of UAV-based techniques can occur. The spatial and spectral resolution of the UAV images, as well as the selection of the appropriate machine learning algorithm, are critical factors in characterizing vegetation [45].
Each individual UAV location serves as a focal point, providing information down to the finest resolution level (e.g., individual plants). Furthermore, a combination of multiple UAS locations can be employed for training satellite data and evaluating vegetation over larger areas. In terms of cost-effectiveness, UAV remote sensing is more suitable for small-scale applications. For example, if the goal is to identify P. kansuensis in an extensive grassland, it would be beneficial to consider adapting the proposed method for future use with higher spatial resolution satellite imagery.
Additionally, the outlined methodological framework holds potential applicability in diverse environments such as alpine regions or rangelands [47]. Spectral information derived from imagery is pivotal in species classification, as distinct spectral characteristics are likely to be exhibited by different species across various wavelength ranges. Therefore, it is crucial to accurately evaluate image quality and process it to extract spectral data (e.g., reflectance). The MS600 Pro sensor (Yusense, Inc., Qingdao, China), which undergoes strict calibration, was utilized for radiometric correction (i.e., converting DNs to reflectance) on the images, and the resulting photographs were processed accordingly. When applying the methodology outlined in the study, it is essential to consider whether relationships identified in one area or year can be extrapolated to other areas and/or years [48]. Images captured by the MS600 Pro camera were taken on clear sunny days, enabling comparisons of reflectance across different fields and dates.
Selecting the appropriate phenological phase based on variations in vegetation phenology can effectively discriminate between different types of vegetation [49], such as distinguishing between dry and rainy seasons [50]. Phenological divergence (e.g., flowering, dormancy) can aid in discerning native plants from invasive species [2,23,51]. The research findings suggest that the flowering period plays a pivotal role in distinguishing P. kansuensis from other plant species. This may be attributed to the influence of bloom mass or visibility on image spectrum, rendering it more readily identifiable with a distinctive appearance. While separating vegetative components (leaves, stems) from surrounding vegetation presents challenges, this is consistent with conclusions drawn from prior investigations into other invasive herbaceous plants [52,53].

4.2. Band and Indices for Overall Accuracy Assessment and Model Validity

Analysis of various combinations of indices, spectral bands, and band + indices revealed a remarkably high level of consensus on P. kansuensis (accuracy: 0.95 to 1) except for the SVM model. These findings suggest that the classifications were robust, even when employing the least effective models tested, underscoring the potential for utilizing UAV imagery in delineating P. kansuensis encroachment and other land types in alpine grasslands. The study demonstrated the potential of UAVs in differentiating between P. kansuensis and other coexisting plant species and land cover in the area. The critical feature selection assessment revealed that the classification primarily relied on the green, red, and red edge bands, which are closely associated with the biochemical properties of vegetation (see Figure 7, Figure 8 and Figure 9). Other similar research confirms the importance of the red edge band in tree type and health status assessments [19], in light of the sensitivity of the red edge band to small variations in the chlorophyll content. In addition to the significance of single bands, ratio and normalized difference vegetation indices—such as B24, B10, B16, B15, B8, B9, B20—that combine these bands also exhibit considerable importance. However, extracting multiple categories based solely on one index is challenging; therefore, it is necessary to utilize multiple indices simultaneously.
Interestingly, it was observed that the bands and indices were only effective for detecting P. kansuensis and did not provide a discernible signal for purple flowering plants (Gentiana scabra, Oxytropis glabra, Astragalus). The distinct color of P. kansuensis makes it easily recognizable, with its long inflorescence being larger than those of Gentiana scabra, Oxytropis glabra, and Astragalus’s, enabling their differentiation from other species at a resolution of 1 cm. Therefore, a 1 cm resolution is sufficient for P. kansuensis but may not be suitable for smaller purple flowering plants due to their size and bloom characteristics, which pose challenges for individual species identification. Additionally, mixed-pixel issues likely contributed to confusion between purple flowering plants, green grass, water bodies and P. kansuensis (refer to Figure 7, Figure 8 and Figure 9). When the spatial resolution is smaller than the size of flowers, spectral bands demonstrated superior performance with an overall accuracy (OA) of 94% except for the SVM model. In contrast, 19 index and spectral band+index combinations exhibited relatively lower performance. Findings of this study align with those of [3,49] who achieved satisfactory classification of woody encroachment using only a 6-band sensor and RF classifier. The strong performance of RGB and red edge images in this study may be attributed to the relatively homogeneous herbaceous vegetation surveyed, which lacks components with reflectance similar to that of flowers (identifiable through red and green bands).
Blooming can also be used to identify and map various invasive plant species, as their vibrant flowers are easily distinguishable in RGB aerial imagery. The majority of the P. kansuensis plant consists of inflorescences during the flowering period, facilitating recognition. Moreover, the size of these inflorescences contributes to accurate validation results when identifying P. kansuensis flowers. Additionally, the spectral characteristics of pixels vary based on the complexity of each species. More complex objects result in greater differences between their spectral features and pure pixel features, leading to reduced accuracy when recognizing scattered patterns [54]. The data presented in Figure 7, Figure 8 and Figure 9 illustrates the values of the training set and plot samples for three plots across two band combinations. Some species’ samples formed clusters with partial overlap, particularly evident in the red vs. green band combinations, where P. kansuensis and green grass clusters were more distinctly separated compared to the red edge vs. infrared combination at all three plots.
The findings of this study are consistent with conclusions drawn from Pedicularis’ spectral curve [39]. Notably, P. kansuensis exhibits a unique spectral curve differing from other grasslands between 550 to 680 nm [55]. During the peak flowering phase of invasive species, there was a notable capacity for distinguishing P. kansuensis from surrounding native species at the pixel level. Water and soil displayed minimal reflectance at red edge 750 nm and NIR band wavelengths. The comparison of blue versus red bands in litter revealed a more pronounced slope value compared to other types in the steppe site. Importantly, discriminating between spectral samples of Oxytropis glabra/Astragalus (purple flowering) and Gentiana scabra (purple flowering) plants using six bands proved challenging due to their diminutive size, posing potential identification difficulties and potential confusion with other grass cover.
The significance of the green and red spectral bands, as well as the vegetation index derived from these bands, is also considered to be of great importance. This could be attributed to the distinct physiological structure of P. kansuensis compared to other types of grass [56]. The analysis revealed that additional green spectral bands, in combination with NIR and SWIR bands, play a significant role in modeling and mapping P. kansuensis. Canopy spectral information related to color can be used for identifying P. kansuensis, which was confirmed by features contributing to high accuracy in species discrimination according to the RF and XGBoost classifiers. The differentiation is more noticeable in meadows than in steppes. Overall, the results indicate a strong potential for using UAV images to map P. kansuensis within various mixed species environments.

5. Conclusions

This study examined the efficacy of UAV characteristics in classifying aggressive P. kansuensis within a diverse grassland environment. The research indicates that six bands and indices can effectively differentiate invasive P. kansuensis and evaluate their flower cover, except for the SVM model. The most significant feature bands for the three field plots, as well as for the invasive species and land cover, were concentrated at 750 nm, 550 nm, and 660 nm. RF and XGBoost yielded precise classification of P. kansuensis, achieving high producer and user accuracies exceeding 95%. These findings suggest that UAV-acquired imagery serves as an unparalleled data source for examining fine-scale grassland species composition due to its high spatial resolution, making UAV a valuable tool for mapping invasive P. kansuensis on grasslands with diverse species backgrounds.
In general, the utilization of UAV spectral bands and vegetation indices has advanced our understanding of P. kansuensis and the surrounding indigenous species, as well as the spectral properties of the soil background. This approach facilitates the exploration of spatial and temporal variations in invasive species coverage and native vegetation, or their response to conservation efforts. It can be particularly valuable for detecting changes in vegetation following invasive species eradication through conservation measures, which may not be discernible using traditional vegetation monitoring methods. However, given the limitations of this approach and the characteristics of grassland ecosystems, it is vital to conduct broader comparisons across larger study areas over extended time periods for multi-temporal analyses. This will bolster conservation endeavors by equipping practitioners with a means to easily monitor and select treatment methods based on target vegetation density (e.g., mechanical or chemical treatments).

Author Contributions

Methodology and software: J.Z. (Jin Zhao) and J.Z. (Jiarong Zhang); Validation: Y.L.; Formal analysis: J.Z. (Jin Zhao); Investigation: J.Z. (Jiarong Zhang) and K.L.; Writing—original draft preparation: J.Z. (Jin Zhao); Writing—review and editing: J.Z. (Jin Zhao) and K.L.; Machine learning: X.L., contributed the same as the corresponding author. All authors have read and agreed to the published version of the manuscript.

Funding

This research was jointly supported by the Third Xinjiang Scientific Expedition Program (Grant No.2021xjkk1400), the National Natural Science Foundation General Program of China (No. 32271747, 42071141) and self-topic Fund of Xinjiang institute of Ecology and Geography, Chinese Academy of Sciences.

Data Availability Statement

The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Bengtsson, J.; Bullock, J.M.; Egoh, B.; Everson, C.; Everson, T.; O’Connor, T.; O’Farrell, P.J.; Smith, H.G.; Lindborg, R. Grasslands-more important for ecosystem services than you might think. Ecosphere 2019, 10, e02582. [Google Scholar] [CrossRef]
  2. Bolch, E.A.; Santos, M.J.; Ade, C.; Khanna, S.; Basinger, N.T.; Reader, M.O.; Hestir, E.L. Remote Detection of Invasive Alien Species. In Remote Sensing of Plant Biodiversity; Cavender-Bares, J., Gamon, J.A., Townsend, P.A., Eds.; Springer Nature: Cham, Switzerland, 2020; pp. 267–307. [Google Scholar]
  3. Bakacsy, L.; Tobak, Z.; van Leeuwen, B.; Szilassi, P.; Biro, C.; Szatmari, J. Drone-Based Identification and Monitoring of Two Invasive Alien Plant Species in Open Sand Grasslands by Six RGB Vegetation Indices. Drones 2023, 7, 207. [Google Scholar] [CrossRef]
  4. Zhao, B.; Liu, Z.; Lu, H.; Wang, Z.; Sun, L.; Wan, X.; Guo, X.; Zhao, Y.; Wang, J.; Shi, Z. Damage and Control of Poisonous Weeds in Western Grassland of China. Agric. Sci. China 2010, 9, 1512–1521. [Google Scholar] [CrossRef]
  5. Biro, E.; Babai, D.; Bodis, J.; Molnar, Z. Lack of knowledge or loss of knowledge? Traditional ecological knowledge of population dynamics of threatened plant species in East-Central Europe. J. Nat. Conserv. 2014, 22, 318–325. [Google Scholar] [CrossRef]
  6. Krug, R.M.; Roura-Pascual, N.; Richardson, D.M. Clearing of invasive alien plants under different budget scenarios: Using a simulation model to test efficiency. Biol. Invasions 2010, 12, 4099–4112. [Google Scholar] [CrossRef]
  7. Shen, S.; Xu, G.; Li, D.; Clements, D.R.; Jin, G.; Yin, X.; Gao, R.; Zhang, F. Occurrence and damage of invasive alien plants in Dehong Prefecture, western of Yunnan Province. Acta Ecol. Sin. 2017, 37, 195–200. [Google Scholar] [CrossRef]
  8. Li, A.; Wu, J.; Zhang, X.; Xue, J.; Liu, Z.; Han, X.; Huang, J. China’s new rural “separating three property rights” land reform results in grassland degradation: Evidence from Inner Mongolia. Land Use Policy 2018, 71, 170–182. [Google Scholar] [CrossRef]
  9. Wijesingha, J.; Astor, T.; Schulze-Brüninghoff, D.; Wachendorf, M. Mapping Invasive Lupinus polyphyllus Lindl. in Semi-Natural Grasslands Using Object-Based Image Analysis of UAV-Borne Images. PFG J. Photogramm. Remote Sens. Geoinf. Sci. 2020, 88, 391–406. [Google Scholar] [CrossRef]
  10. Zhao, Y.; Zhu, W.; Wei, P.; Fang, P.; Zhang, X.; Yan, N.; Liu, W.; Zhao, H.; Wu, Q. Classification of Zambian grasslands using random forest feature importance selection during the optimal phenological period. Ecol. Indic 2022, 135, 108529. [Google Scholar] [CrossRef]
  11. Wang, D.; Cui, B.; Duan, S.; Chen, J.; Fan, H.; Lu, B.; Zheng, J. Moving north in China: The habitat of Pedicularis kansuensis in the context of climate change. Sci. Total Environ. 2019, 697, 133979. [Google Scholar] [CrossRef]
  12. Yu, J.; Hu, Y.; Li, K.; Ma, X. Study on the biological characteristics of the medicinal plant of the Tianshan Mountain. Chin. Tradit. Herb. Drugs 2006, 37, 1884–1886. [Google Scholar]
  13. Liu, Y.; Hu, Y.; Yu, J.; Li, K.; Gao, G.; Wang, X. Study on harmfulness of Pedicularis myriophylla and its control measures. Arid Zone Res 2008, 25, 778–782. [Google Scholar] [CrossRef]
  14. Liu, M.; Dries, L.; Heijman, W.; Zhu, X.; Deng, X.; Huang, J. Land tenure reform and grassland degradation in Inner Mongolia, China. China Econ. Rev. 2019, 55, 181–198. [Google Scholar] [CrossRef]
  15. Mullerova, J.; Bruna, J.; Bartalos, T.; Dvorak, P.; Vitkova, M.; Pysek, P. Timing Is Important: Unmanned Aircraft vs. Satellite Imagery in Plant Invasion Monitoring. Front. Plant Sci. 2017, 8, 887. [Google Scholar] [CrossRef]
  16. Huang, C.; Geiger, E.L.; Van Leeuwen, W.J.D.; Marsh, S.E. Discrimination of invaded and native species sites in a semi-desert grassland using MODIS multi-temporal data. Int. J. Remote Sens. 2009, 30, 897–917. [Google Scholar] [CrossRef]
  17. Sun, Y.; Yi, S.; Hou, F. Unmanned aerial vehicle methods makes species composition monitoring easier in grasslands. Ecol. Indic. 2018, 95, 825–830. [Google Scholar] [CrossRef]
  18. Lopatin, J.; Fassnacht, F.E.; Kattenborn, T.; Schmidtlein, S. Mapping plant species in mixed grassland communities using close range imaging spectroscopy. Remote Sens. Environ 2017, 201, 12–23. [Google Scholar] [CrossRef]
  19. Abdollahnejad, A.; Panagiotidis, D. Tree Species Classification and Health Status Assessment for a Mixed Broadleaf-Conifer Forest with UAS Multispectral Imaging. Remote Sens. 2020, 12, 3722. [Google Scholar] [CrossRef]
  20. Rango, A.; Laliberte, A.; Herrick, J.E.; Winters, C.; Havstad, K.; Steele, C.; Browning, D. Unmanned aerial vehicle-based remote sensing for rangeland assessment, monitoring, and management. J. Appl. Remote Sens 2009, 3, 033542. [Google Scholar] [CrossRef]
  21. Lyu, X.; Li, X.; Dang, D.; Dou, H.; Wang, K.; Lou, A. Unmanned Aerial Vehicle (UAV) Remote Sensing in Grassland Ecosystem Monitoring: A Systematic Review. Remote Sens. 2022, 14, 1096. [Google Scholar] [CrossRef]
  22. De Sa, N.C.; Castro, P.; Carvalho, S.; Marchante, E.; Lopez-Nunez, F.A.; Marchante, H. Mapping the Flowering of an Invasive Plant Using Unmanned Aerial Vehicles: Is There Potential for Biocontrol Monitoring? Front. Plant Sci. 2018, 9, 293. [Google Scholar] [CrossRef] [PubMed]
  23. Martin, F.-M.; Mullerova, J.; Borgniet, L.; Dommanget, F.; Breton, V.; Evette, A. Using Single- and Multi-Date UAV and Satellite Imagery to Accurately Monitor Invasive Knotweed Species. Remote Sens. 2018, 10, 1662. [Google Scholar] [CrossRef]
  24. Abeysinghe, T.; Milas, A.S.; Arend, K.; Hohman, B.; Reil, P.; Gregory, A.; Vazquez-Ortega, A. Mapping Invasive Phragmites australis in the Old Woman Creek Estuary Using UAV Remote Sensing and Machine Learning Classifiers. Remote Sens. 2019, 11, 1380. [Google Scholar] [CrossRef]
  25. Li, J.; Li, D.; Zhang, G.; Xu, H.; Zeng, R.; Luo, W.; Yu, Y. Study on extraction of foreign invasive species Mikania micrantha based on unmanned aerial vehicle (UAV) hyperspectral remote sensing. In Proceedings of the 5th Symposium on Novel Optoelectronic Detection Technology and Application, Xi’an, China, 24–26 October 2019; p. 53. [Google Scholar]
  26. Schmidt, J.; Fassnacht, F.E.; Neff, C.; Lausch, A.; Kleinschmit, B.; Foerster, M.; Schmidtlein, S. Adapting a Natura 2000 field guideline for a remote sensing-based assessment of heathland conservation status. Int. J. Appl. Earth Obs. Geoinf. 2017, 60, 61–71. [Google Scholar] [CrossRef]
  27. Lu, B.; He, Y. Species classification using Unmanned Aerial Vehicle (UAV)-acquired high spatial resolution imagery in a heterogeneous grassland. Int. J. Photogramm. Remote Sens. 2017, 128, 73–85. [Google Scholar] [CrossRef]
  28. Meng, B.; Yang, Z.; Yu, H.; Qin, Y.; Sun, Y.; Zhang, J.; Chen, J.; Wang, Z.; Zhang, W.; Li, M.; et al. Mapping of Kobresia pygmaea Community Based on Umanned Aerial Vehicle Technology and Gaofen Remote Sensing Data in Alpine Meadow Grassland: A Case Study in Eastern of Qinghai-Tibetan Plateau. Remote Sens. 2021, 13, 2483. [Google Scholar] [CrossRef]
  29. Yang, H.; Du, J. Classification of desert steppe species based on unmanned aerial vehicle hyperspectral remote sensing and continuum removal vegetation indices. Optik 2021, 247, 167877. [Google Scholar] [CrossRef]
  30. Masemola, C.; Cho, M.A.; Ramoelo, A. Sentinel-2 time series based optimal features and time window for mapping invasive Australian native Acacia species in KwaZulu Natal, South Africa Cecilia. Int. J. Appl. Earth Obs. Geoinf. 2020, 93, 102207. [Google Scholar] [CrossRef]
  31. Ollinger, S.V. Sources of variability in canopy reflectance and the convergent properties of plants. New Phytol. 2011, 189, 375–394. [Google Scholar] [CrossRef]
  32. Bradley, B.A. Remote detection of invasive plants: A review of spectral, textural and phenological approaches. Biol. Invasions 2014, 16, 1411–1425. [Google Scholar] [CrossRef]
  33. Skowronek, S.; Ewald, M.; Isermann, M.; Van de Kerchove, R.; Lenoir, J.; Aerts, R.; Warrie, J.; Hattab, T.; Honnay, O.; Schmidtlein, S.; et al. Mapping an invasive bryophyte species using hyperspectral remote sensing data. Biol. Invasions 2017, 19, 239–254. [Google Scholar] [CrossRef]
  34. Somers, B.; Asner, G.P. Multi-temporal hyperspectral mixture analysis and feature selection for invasive species mapping in rainforests. Remote Sens. Environ. 2013, 136, 14–27. [Google Scholar] [CrossRef]
  35. Gioria, M.; Pysek, P.; Osborne, B.A. Timing is everything: Does early and late germination favor invasions by herbaceous alien plants? J. Plant Ecol. 2018, 11, 4–16. [Google Scholar] [CrossRef]
  36. Somodi, I.; Carni, A.; Ribeiro, D.; Podobnikar, T. Recognition of the invasive species Robinia pseudacacia from combined remote sensing and GIS sources. Biol. Conserv. 2012, 150, 59–67. [Google Scholar] [CrossRef]
  37. Mirik, M.; Chaudhuri, S.; Surber, B.; Ale, S.; Ansley, R.J. Detection of two intermixed invasive woody species using color infrared aerial imagery and the support vector machine classifier. J. Appl. Remote Sens. 2013, 7, 073588. [Google Scholar] [CrossRef]
  38. Laliberte, A.S.; Rango, A.; Havstad, K.M.; Paris, J.F.; Beck, R.F.; McNeely, R.; Gonzalez, A.L. Object-oriented image analysis for mapping shrub encroachment from 1937 to 2003 in southern New Mexico. Remote Sens. Environ. 2004, 93, 198–210. [Google Scholar] [CrossRef]
  39. Yang, H. Comparative Study of Object-Oriented Forest Classification Methods Considering Feature Type and Feature Selection. Master’s Thesis, Shaanxi Normal University, Xi’an, China, 2019. [Google Scholar]
  40. Wang, W.; Tang, J.; Zhang, N.; Xu, X.; Zhang, A.; Wang, Y. Automated Detection Method to Extract Pedicularis Based on UAV Images. Drones 2022, 6, 399. [Google Scholar] [CrossRef]
  41. Treccani, D.; Adami, A.; Fregonese, L. Drones and Real-Time Kinematic Base Station Integration for Documenting Inaccessible Ruins: A Case Study Approach. Drones. 2024, 8, 268. [Google Scholar] [CrossRef]
  42. Shao, Z.; Ahmad, M.N.; Javed, A. Comparison of Random Forest and XGBoost Classifiers Using Integrated Optical and SAR Features for Mapping Urban Impervious Surface. Remote Sens. 2024, 16, 665. [Google Scholar] [CrossRef]
  43. Bartold, M.; Kluczek, M. A Machine Learning Approach for Mapping Chlorophyll Fluorescence at Inland Wetlands. Remote Sens. 2023, 15, 2392. [Google Scholar] [CrossRef]
  44. Costa, L.S.; Sano, E.E.; Ferreira, M.E.; Munhoz, C.B.R.; Costa, J.V.S.; Alves Junior, L.R.; de Mello, T.R.B.; Bustamante, M.M.d.C. Woody Plant Encroachment in a Seasonal Tropical Savanna: Lessons about Classifiers and Accuracy from UAV Images. Remote Sens. 2023, 15, 2342. [Google Scholar] [CrossRef]
  45. Torresani, M.; Kleijn, D.; de Vries, J.P.R.; Bartholomeus, H.; Chieffallo, L.; Gatti, R.C.; Moudry, V.; Da Re, D.; Tomelleri, E.; Rocchini, D. A novel approach for surveying flowers as a proxy for bee pollinators using drone images. Ecol. Indic. 2023, 149, 110123. [Google Scholar] [CrossRef]
  46. Gonzales, D.; de Ibarra, N.H.; Anderson, K. Remote Sensing of Floral Resources for Pollinators—New Horizons from Satellites to Drones. Front. Ecol. Evol. 2022, 10, 869751. [Google Scholar] [CrossRef]
  47. Riihimaki, H.; Luoto, M.; Heiskanen, J. Estimating fractional cover of tundra vegetation at multiple scales using unmanned aerial systems and optical satellite data. Remote Sens. Environ. 2019, 224, 119–132. [Google Scholar] [CrossRef]
  48. Blasi, M.; Bartomeus, I.; Bommarco, R.; Gagic, V.; Garratt, M.; Holzschuh, A.; Kleijn, D.; Lindstrom, S.A.M.; Olsson, P.; Polce, C.; et al. Evaluating predictive performance of statistical models explaining wild bee abundance in a mass-flowering crop. Ecography 2021, 44, 525–536. [Google Scholar] [CrossRef]
  49. Zhang, X.; Xiao, X.; Wang, X.; Xu, X.; Chen, B.; Wang, J.; Ma, J.; Bin, Z.; Li, B. Quantifying expansion and removal of Spartina alterniflora on Chongming island, China, using time series Landsat images during 1995–2018. Remote Sens. Environ. 2020, 247, 111916. [Google Scholar] [CrossRef]
  50. Wu, N.; Shi, R.; Zhuo, W.; Zhang, C.; Zhou, B.; Xia, Z.; Tao, Z.; Gao, W.; Tian, B. A Classification of Tidal Flat Wetland Vegetation Combining Phenological Features with Google Earth Engine. Remote Sens. 2021, 13, 443. [Google Scholar] [CrossRef]
  51. Niphadkar, M.; Nagendra, H. Remote Sensing of Invasive Plants: Incorporating Functional Traits into the Picture. Int. J. Remote Sens. 2016, 37, 3074–3085. [Google Scholar] [CrossRef]
  52. Graenzig, T.; Fassnacht, F.E.; Kleinschmit, B.; Foerster, M. Mapping the fractional coverage of the invasive shrub Ulex europaeus with multi-temporal Sentinel-2 imagery utilizing UAV orthoimages and a new spatial optimization approach. Int. J. Appl. Earth Obs. Geoinf. 2021, 96, 102281. [Google Scholar] [CrossRef]
  53. Altamirano, A.; Paola Cely, J.; Etter, A.; Miranda, A.; Fuentes-Ramirez, A.; Acevedo, P.; Salas, C.; Vargas, R. The invasive species Ulex europaeus (Fabaceae) shows high dynamism in a fragmented landscape of south-central Chile. Environ. Monit. Assess. 2016, 188, 495. [Google Scholar] [CrossRef]
  54. Wang, W.; Tang, J.; Zhang, N.; Wang, Y.; Xu, X.; Zhang, A. Spatiotemporal Pattern of Invasive Pedicularis in the Bayinbuluke Land, China, during 2019-2021: An Analysis Based on PlanetScope and Sentinel-2 Data. Remote Sens. 2023, 15, 4383. [Google Scholar] [CrossRef]
  55. Gao, S.; Lin, J.; Ma, T.; Wu, J.; Zheng, J. Extraction and Analysis of Hyperspectral Data and Characteristics from Pedicularis on Bayanbulak Grassland in Xinjiang. Remote Sens. Technol. Appl. 2018, 33, 908–914. [Google Scholar]
  56. Hu, J.D.; Li, K.H.; Deng, C.J.; Gong, Y.M.; Liu, Y.Y.; Wang, L. Seed Germination Ecology of Semiparasitic Weed Pedicularis kansuensis in Alpine Grasslands. Plants 2022, 11, 1777. [Google Scholar] [CrossRef] [PubMed]
Drones 08 00639 g001
Figure 1. Location of UAV sites.
Figure 1. Location of UAV sites.
Drones 08 00639 g001
Drones 08 00639 g002
Figure 2. Flowchart of classification of invasive species and grassland.
Figure 2. Flowchart of classification of invasive species and grassland.
Drones 08 00639 g002
Drones 08 00639 g003
Figure 3. Orthoimage and field photo of the quadrat.
Figure 3. Orthoimage and field photo of the quadrat.
Drones 08 00639 g003
Drones 08 00639 g004
Figure 4. Classification map based on the spectral bands of UAV orthoimages.
Figure 4. Classification map based on the spectral bands of UAV orthoimages.
Drones 08 00639 g004
Drones 08 00639 g005aDrones 08 00639 g005b
Figure 5. Accuracy assessment for UAV orthoimages. The color ranges from 0 to 1, with red indicating lower accuracy and white indicating higher accuracy.
Figure 5. Accuracy assessment for UAV orthoimages. The color ranges from 0 to 1, with red indicating lower accuracy and white indicating higher accuracy.
Drones 08 00639 g005aDrones 08 00639 g005b
Drones 08 00639 g006
Figure 6. Feature importance of spectral bands and indices.
Figure 6. Feature importance of spectral bands and indices.
Drones 08 00639 g006
Drones 08 00639 g007
Figure 7. Scatter plots of two bands in swamp meadow.
Figure 7. Scatter plots of two bands in swamp meadow.
Drones 08 00639 g007
Drones 08 00639 g008
Figure 8. Scatter plots of two bands in alpine steppe1.
Figure 8. Scatter plots of two bands in alpine steppe1.
Drones 08 00639 g008
Drones 08 00639 g009
Figure 9. Scatter plots of two bands in alpine steppe2.
Figure 9. Scatter plots of two bands in alpine steppe2.
Drones 08 00639 g009
Table 1. Formula for the spectral band and indices used in this study.
Table 1. Formula for the spectral band and indices used in this study.
Feature CategoryBand NameFeatureFormula and Wavelength
Spectral bandB1BlueR450 *
B2GreenR555
B3RedR660
B4Rededge720 R720
B5Rededge750R750
B6NIRR840
Characteristics indexB7RENDVI720(R720 − R660)/(R720 + R660)
B8RENDVI750 (R750 − R660)/(R750 + R660)
B9NDVI(R840 − R660)/(R840 + R660)
B10NDRG(R660 − R555)/(R660 + R555)
B11NDGB(R555 − R450)/(R555 + R450)
B12NDBG(R450 − R555)/(R555 + R450)
B13NDRB(R660 − R450)/(R660 + R450)
B14NDBR(R450 − R660)/(R660 + R450)
B15V4/PI × arctan((R555 − R450)/(R555 + R450))
B16S4/PI × arctan(1 − (R555 − R450)/(R555 + R450))
B17GNDVI(R840 − R555)/(R840 + R555)
B18GRVIR840/R555
B19MTCI720(R840 − R720)/(R840 + R720)
B20MTCI750(R840 − R750)/(R840 + R750)
B21CI720(R840 − R720)/R720
B22CI750(R840 − R750)/R750
B23EVI2.5 × (R840 − R660)/(R840 + R660 + 1)
B24SR-RGR660/R555
B25SR-GBR555/R450
* ”R” stands for reflectance, and the numbers denote the band position. The unit of the band is nm.
Table 2. Grasslands, main communities, and other land cover types in UAV images of three plots.
Table 2. Grasslands, main communities, and other land cover types in UAV images of three plots.
GrasslandMain CommunitiesFlowering Stage SpeciesLand CoverDate
Alpine steppe1Stipa purpurea, Festuca ovina, Agropyron cristatum, P. kansuensisP. kansuensisLitter, Soil Grass (green),2 August 2022
Alpine steppe2Carex turkestanica, Festuca kryloviana, P. kansuensisGentiana scabra (purple), P. kansuensisGrass (green), Soil, Water2 August 2022
Swamp meadowCarex stipitiutriculata, P. kansuensisOxytropis glabra/Astragalus (purple), P. kansuensis, Umbelliferae (white)Grass (green), Soil, Water3 August 2022
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Zhao, J.; Li, K.; Zhang, J.; Liu, Y.; Li, X. Mapping Invasive Species Pedicularis and Background Grassland Using UAV and Machine Learning Algorithms. Drones 2024, 8, 639. https://doi.org/10.3390/drones8110639

AMA Style

Zhao J, Li K, Zhang J, Liu Y, Li X. Mapping Invasive Species Pedicularis and Background Grassland Using UAV and Machine Learning Algorithms. Drones. 2024; 8(11):639. https://doi.org/10.3390/drones8110639

Chicago/Turabian Style

Zhao, Jin, Kaihui Li, Jiarong Zhang, Yanyan Liu, and Xuan Li. 2024. "Mapping Invasive Species Pedicularis and Background Grassland Using UAV and Machine Learning Algorithms" Drones 8, no. 11: 639. https://doi.org/10.3390/drones8110639

APA Style

Zhao, J., Li, K., Zhang, J., Liu, Y., & Li, X. (2024). Mapping Invasive Species Pedicularis and Background Grassland Using UAV and Machine Learning Algorithms. Drones, 8(11), 639. https://doi.org/10.3390/drones8110639

Article Metrics

Back to TopTop