Next Article in Journal
Precision Feeding in Lactating Sows Improves Growth Performance and Carcass Quality of Their Progeny
Previous Article in Journal
Measurement of Production Efficiency and Analysis of Influencing Factors in Major Sugarcane-Producing Regions of China
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Analytical Methods for Wind-Driven Dynamic Behavior of Pear Leaves (Pyrus pyrifolia)

1
School of Agricultural Engineering, Jiangsu University, Zhenjiang 212013, China
2
Key Laboratory of Plant Protection Engineering, Ministry of Agriculture and Rural Affairs, Jiangsu University, Zhenjiang 212013, China
*
Author to whom correspondence should be addressed.
Agriculture 2025, 15(8), 886; https://doi.org/10.3390/agriculture15080886
Submission received: 13 March 2025 / Revised: 5 April 2025 / Accepted: 17 April 2025 / Published: 18 April 2025
(This article belongs to the Section Agricultural Technology)

Abstract

:
The fluttering of leaves under wind fields significantly impacts the efficiency and precision of agricultural spraying. However, existing spraying technologies often overlook the complex mechanisms of wind–leaf interactions. This study integrates the fine-tuned Segment Anything Model 2 with multi-dimensional dynamic behavior analysis to provide a systematic approach for investigating leaf fluttering under wind fields. First, a segmentation algorithm based on Principal Component Analysis was employed to eliminate background interference in leaf fluttering data. The results showed that the segmentation algorithm achieved an Intersection over Union (IoU) ranging from 98.2% to 98.7%, with Precision reaching 99.0% to 99.5%, demonstrating high segmentation accuracy and reliability. Building on this, experiments on leaf segmentation and tracking in dynamic scenarios were conducted using the SAM2-FT model. The results indicated that SAM2-FT effectively captured the dynamic behavior of leaves by integrating spatiotemporal information, achieving Precision and AP50/% values exceeding 97%. Its overall performance significantly outperformed mainstream YOLO-series models. In the analysis of dynamic response patterns, the Hilbert transform and time-series quantification methods were introduced to reveal the amplitude, frequency, and trajectory characteristics of a leaf fluttering under wind fields across three dimensions: area, inclination angle, and centroid. This comprehensive analysis highlights the dynamic response characteristics of leaves to wind field perturbations.

1. Introduction

In orchard management, spraying pesticides is a critical step for controlling pests and diseases, with its efficiency and precision being pivotal to improving pesticide utilization and environmental protection [1,2]. In recent years, air-assisted spraying technology has been widely applied in orchard protection operations because of its high efficiency and uniform coverage [3,4]. This technology enhances the penetration of droplets through the tree canopy and significantly improves pesticide deposition rates by leveraging wind fields [5,6,7]. However, variations in wind fields substantially influence droplet trajectories and the dynamic behavior of leaves during spraying processes [8,9,10]. In real-world scenarios, wind-induced leaf fluttering and morphological changes introduce considerable uncertainties to spraying applications, reducing droplet coverage while leading to pesticide waste and environmental pollution [10,11].
Existing studies have identified wind fields as a key external factor influencing spray distribution [12,13]. Excessive wind speeds or unstable wind directions can significantly alter droplet trajectories, resulting in spray drift and reduced coverage [14,15,16]. Meanwhile, leaves exhibit notable nonlinear motion characteristics under wind disturbances, such as rotation, fluttering, and bending. These dynamic behaviors further change the posture and spatial distribution of target leaves, affecting droplet deposition efficacy [17,18,19]. However, traditional spray optimization research primarily focuses on droplet distribution under static leaf conditions [20], paying little attention to the multifaceted impacts of wind–leaf interactions in dynamic environments.
In recent years, growing attention has been paid to the response characteristics of leaves in dynamic environments, particularly the effects of wind fields on leaf dynamics. Computational Fluid Dynamics (CFD) simulation has emerged as one of the most commonly used methods to explore fluid dynamics and leaf response mechanisms under wind disturbances [21,22,23]. Significant progress has been achieved in revealing the fluid dynamic characteristics and response mechanisms of leaves under wind disturbances using CFD [23,24]. Moreover, the combination of CFD and Finite Element Method (FEM) has been employed to simulate the coupled interactions between wind fields and leaves [25,26], uncovering the nonlinear response characteristics of leaves under specific wind conditions, such as twisting and severe deformation at high wind speeds. Despite the theoretical insights provided by these methods, their reliance on idealized boundary conditions and preset parameters makes it challenging to fully capture the realistic dynamic characteristics of leaves in complex natural scenarios. Additionally, the high computational cost of CFD simulations limits their application to multi-leaf systems and large-scale dynamic scenarios. To address these limitations, techniques leveraging real-world dynamic data have become an essential supplement for studying leaf dynamics, offering novel approaches to understanding wind–leaf interaction mechanisms.
With the rapid advancement of computer vision and deep learning technologies, these methods have been increasingly utilized in agricultural spray research [27,28,29], providing novel insights into wind–leaf interaction mechanisms. Traditional computer vision techniques for analyzing leaf motion typically rely on high-speed imaging, often coupled with feature point tracking, to capture dynamic behavior. However, such methods primarily target local key points, such as leaf tips, while overlooking the coordinated variations in overall morphology, posture, and spatial position. Consequently, they lack the capacity to comprehensively characterize the complex dynamic responses of leaves under wind disturbances.
To address these limitations, this study focuses on the dynamic response behavior of leaves under controlled wind field conditions and proposes a systematic framework that integrates a fine-tuned Segment Anything Model 2 (SAM2-FT) with multi-dimensional dynamic behavior analysis. The methodology involves the following steps:
  • Real-world leaf fluttering video data were acquired under laboratory-controlled wind conditions at the Yangyu Ecological Pear Orchard in Jiangsu Province, ensuring both representativeness and experimental repeatability;
  • A large-scale, high-quality dynamic leaf dataset was constructed using a color segmentation technique based on Principal Component Analysis (PCA);
  • A SAM2-FT-based segmentation and tracking model was developed for precise leaf detection and temporal tracking in dynamic scenes. Furthermore, Hilbert transform and time-series analysis were applied to quantify leaf dynamics across multiple dimensions, including area, inclination angle, and centroid position.
This approach enables the accurate capture and multi-scale quantification of leaf dynamic responses, providing a novel methodological foundation for investigating wind–leaf interaction mechanisms.

2. Materials and Methods

2.1. Experimental Data Collection

The pear leaf image samples used in this study were collected from the Yangyu Ecological Pear Orchard in Jiangsu Province. The orchard employs a trellis planting system with a triangular layout, featuring a row spacing of 5 m, plant spacing of 3 m, and a trellis height of 3 m, all in accordance with standardized orchard planting practices (Figure 1). Under laboratory-controlled conditions, wind-induced leaf fluttering data were recorded using an action camera at a fixed wind speed. A total of 10 sets of leaf fluttering video data were collected, each lasting 20 s, providing a reliable dataset for analyzing the fluttering patterns of leaves.

2.2. Experimental Data Processing

Data processing involved frame-by-frame background separation of the videos and dataset construction to facilitate subsequent model fine-tuning and reduce uncertainties in the analysis of leaf fluttering. In this study, color features were captured in the HSV color space by analyzing the HSV thresholds of leaf images [30]. A color segmentation technique was applied to separate the background from the fluttering leaf images on a frame-by-frame basis in the videos.
Leveraging the three-dimensional characteristics of the HSV color space, Principal Component Analysis (PCA) [31,32] was employed to reduce the dimensionality of leaf pixel information while retaining key color features (Figure 2). PCA extracts two principal components to reduce data dimensionality while preserving the most discriminative color characteristics. In the reduced two-dimensional PCA space, random sampling was applied to ensure data representativeness and reduce computational complexity. This study utilized the Minibatch K-Means algorithm [33] to cluster color data, classifying pixels into two categories: background and leaf. The HSV ranges for each cluster were then calculated to determine segmentation thresholds, enabling precise color segmentation.
For subsequent fine-tuning of the segmentation model, the LabelMe tool was used to annotate the leaf images in video frames accurately. A total of 2800 leaf mask images were generated (Figure 3), enhancing the recognition of fluttering leaf features and ensuring the model maintains high accuracy and robustness in dynamic and deformed leaf scenarios.

2.3. Analysis of Leaf Fluttering Patterns

2.3.1. SAM2 Model for Detecting Fluttering Leaves with Adaptive Fine-Tuning

To analyze the fluttering patterns of leaves, this study employs the Segment Anything Model 2 (SAM2) as the detection model. SAM2 has been pre-trained on large-scale, diverse image and video datasets, enabling it to adapt to new tasks in unseen scenarios rapidly. By integrating temporal information, the model constructs a motion prediction mechanism for targets, leveraging feature correlations between consecutive frames. This approach achieves high-precision boundary localization of leaves in the spatial dimension while ensuring stable and consistent target tracking in the temporal dimension.
In dynamic and deformed leaf fluttering scenarios, leaves undergo continuous morphological changes during motion, such as bending, rotation, and jitter, which reflect the unique features and constraints of these conditions. By utilizing 2800 annotated leaf mask images from video frames (Figure 3), the model underwent adaptive fine-tuning to optimize key parameters locally, achieving a balance between generalization capability and task-specific adaptability.
The prompting mechanism of the adaptively fine-tuned Segment Anything Model 2 (SAM2-FT) supports various annotation methods for fluttering leaf videos, including points, bounding boxes, or region masks. In this study, key points were used to annotate fluttering leaves in the initial frame. During frame-by-frame video processing, the model employed its robust feature extraction and temporal modeling capabilities to segment and analyze the fluttering leaf regions in each frame. Through a memory attention mechanism, SAM2-FT maintained consistent cross-frame tracking and dynamic adjustments, precisely capturing boundary changes and displacement trajectories of leaves in dynamic scenarios (Figure 4).
The SAM2-FT model enabled accurate acquisition of leaf motion information, providing a reliable data foundation for subsequent analyses of leaf fluttering patterns.

2.3.2. Multi-Dimensional Dynamic Behavior Analysis of Leaves

To further investigate the fluttering patterns of leaves, this study utilizes the SAM2-FT model to extract time-series data on the area, centroid, and angle of fluttering leaves (Figure 5). These metrics comprehensively characterize the dynamic properties of a leaf fluttering across three dimensions—morphology, position, and orientation—thereby elucidating the mechanisms by which leaves respond to environmental changes.
Amplitude and rate analyses are introduced to quantify the fluctuation range of leaf area changes, thereby revealing the intensity, regularity, speed, and dynamic trends of leaf fluttering characteristics. These analyses reflect the sensitivity and response of leaves to external environmental changes, such as variations in wind speed. The definitions are expressed by Equations (1) and (2).
A a m p , k = m a x t [ t k 1 , t k ] A ( t ) m i n t [ t k 1 , t k ] A ( t )
R L e a f ( t i ) = A L e a f t A L e a f ( t i + 1 ) A L e a f ( t i 1 ) t i + 1 t i 1
In the equations, Aamp,k represents the amplitude of the k-th cycle, while maxA(t) and minA(t) denote the peak and trough values of the leaf area within the given time interval, respectively. RLeaf(ti) represents the rate of leaf area change at time ti, where ALeaf (ti + 1) and ALeaf (ti − 1) correspond to the leaf area values at times ti + 1 and ti − 1, respectively.
The Hilbert transform is utilized to calculate the amplitude envelope and instantaneous frequency of the leaf angle signals, uncovering the dynamic intensity and frequency characteristics of leaf fluttering from both frequency-domain and time-domain perspectives. By analyzing the amplitude variations of angle, the rotational amplitude and posture adjustment capability of leaves under external wind fields are quantified. Simultaneously, the instantaneous frequency captures the rate of angle changes and frequency fluctuations at different time points, reflecting the mechanism by which leaves respond to external environmental changes, such as variations in wind speed. The equations are defined as Equations (3) and (4).
E ( t ) = z ( t ) = x ( t ) 2 + x H i b e r t ( t ) 2
f ( t ) = 1 2 π d d t tan 1 x H i b e r t ( t ) x ( t )
In the equations, E(t) and f(t) represent the instantaneous amplitude and frequency of the angle at each time point t, respectively. z(t), x(t), and xHilbert(t) denote the analytic signal, the original signal, and the Hilbert-transformed signal, respectively.
Monitoring the position and instantaneous velocity changes in the leaf centroid on a two-dimensional plane quantifies the spatial characteristics of overall leaf motion. This approach provides a comprehensive description of the range, direction, and dynamic patterns of leaf movement. It reflects the influence of external environmental factors, such as wind speed, on leaf motion trajectories, offering a direct basis for studying leaf fluttering behavior and dynamic interactions. The equations are defined as Equation (5).
v ( t ) = x ( t i + 1 ) x ( t i ) t i + 1 t i 2 + y ( t i + 1 ) y ( t i ) t i + 1 t i 2
In the equation, v(t) represents the instantaneous velocity of the centroid.
This method enables the quantitative characterization of key dynamic features of leaf fluttering, revealing the variation patterns of area fluctuations, angle adjustments, and centroid motion. It systematically reflects the spatial, postural, and morphological characteristics of leaf fluttering, as well as its dynamic response mechanisms and regularities under the influence of wind fields.

3. Results and Discussion

3.1. SAM2-FT Segmentation and Background Separation Experiments

3.1.1. Comparison of Model Segmentation Performance

Accurate target segmentation is a prerequisite for subsequent video tracking. To validate the superiority of SAM2-FT in leaf segmentation tasks, a comparative experiment was conducted against mainstream YOLO segmentation models, including YOLOv8x-seg, YOLOv9e-seg, and YOLOv10x-seg. Precision, Recall, AP50, and F1-score were introduced as evaluation metrics to analyze the segmentation results quantitatively.
A total of 1000 video frame leaf images were selected as segmentation test samples (sample images shown in Figure 6), and the same dataset was used for adaptive fine-tuning of SAM2 and training of YOLO series models. The dataset comprised 2800 video frames of fluttering leaf images, and segmentation experiments were conducted under consistent test conditions.
The experimental environment was configured with the Windows 11 operating system, an i7-14700F CPU, and an NVIDIA GeForce RTX 4060Ti GPU (Lenovo, Beijing, China). The results of the experiments are presented in Table 1.
Experimental data (as shown in Table 1) demonstrate that SAM2-FT significantly outperforms YOLO series segmentation models, including YOLOv8x-seg, YOLOv9e-seg, and YOLOv10x-seg, in leaf segmentation tasks. In terms of Precision, SAM2-FT achieves 98.7%, a notable improvement compared with the highest value among YOLO models (89.56%), enabling highly accurate leaf segmentation (Figure 7 and Video S1). For Recall, SAM2-FT reaches 97.48%, significantly surpassing the highest value of 85.15% from the YOLO series, effectively reducing the miss rate. Additionally, SAM2-FT achieves AP50 and F1 scores of 98.56% and 98.08%, respectively, both substantially higher than those of YOLO models, showcasing its outstanding overall performance.
The core advantages of the SAM2-FT model over mainstream YOLO series models lie in its flexible prompting mechanism, which uses point prompts to mark the initial leaf positions in the first frame. This approach effectively integrates local features with global context, ensuring high segmentation precision and frame-to-frame continuity. Furthermore, SAM2’s pretraining on large-scale, diverse datasets endows it with robust feature extraction and generalization capabilities, while its multi-head self-attention mechanism further enhances segmentation accuracy. Leveraging these advantages in prompting mechanisms and architectural design, SAM2-FT demonstrates exceptional performance in dynamic leaf segmentation tasks.

3.1.2. Background Separation Experiment

A background segmentation experiment for fluttering leaf videos was conducted using 10 sets of collected fluttering leaf video data as test samples. Video data were first converted into video frame images, and 50 sample images were randomly selected (Figure 8). Subsequently, PCA analysis was applied to evaluate the HSV thresholds used for color segmentation, aiming to determine the optimal HSV thresholds for accurate segmentation.
The results of PCA color analysis are presented in Figure 9. Leaf pixel information was reduced to a two-dimensional principal component space, where data points in the 2D PCA space exhibited clear clustering characteristics, reducing dimensionality while retaining the critical color features of the background and leaves. In the reduced 2D PCA space, the clustering feature map of Minibatch K-Means clearly demonstrated the separation between two clusters: Cluster 0 (background) and Cluster 1 (leaf). The clustering results revealed the distribution characteristics of Hue and Saturation, as well as Hue and Value, for leaves and the background.
By applying PCA to reduce the dimensionality of HSV color space data, the complexity of these data was simplified while preserving key color features. Through clustering analysis and threshold extraction, the HSV color information of leaves and background was accurately separated.
The color heatmap (Figure 10) illustrates the HSV threshold distribution range across 50 images. Based on the intersection analysis of all samples, the optimal segmentation thresholds for leaves were determined as follows: Hue: 35–50, Saturation: 56–246, and Value: 33–171. These thresholds precisely encapsulate the key color features of the leaves.
The optimal segmentation thresholds for leaves were applied to perform frame-by-frame background segmentation on fluttering leaf video data. To ensure the reliability of the experimental results, the 10 video datasets were divided into five groups, named leafV_Seg1, leafV_Seg2, leafV_Seg3, leafV_Seg4, and leafV_Seg5, with each group containing two video datasets.
In this experiment, IoU, Precision, Recall, and F1-score were introduced as evaluation metrics. A statistical analysis of the frame-by-frame segmentation results for each group of videos (Figure 11) was conducted to assess the performance and stability of the segmentation algorithm comprehensively.
Experimental results (as shown in Table 2) indicate that the segmentation algorithm demonstrates both reliability and stability in the task of separating leaves from the background. All metrics (IoU, Precision, Recall, and F1-score) exhibited a variation range of less than 1%, reflecting consistent segmentation performance. IoU remained stable between 98.2% and 98.7%, and Precision reached 99.0% to 99.5%, indicating complete segmentation regions with minimal misclassification. Recall ranged from 98.0% to 98.6%, with minor missed detections observed primarily in leaf edge areas affected by lesions, where the color characteristics significantly differ from typical leaf features, making them more challenging to separate fully.
The consistent segmentation performance across all data groups indicates that the algorithm is capable of adapting to segmentation tasks in dynamic leaf scenarios, thereby achieving precise background segmentation in fluttering leaf videos.

3.2. Experiment on Multi-Dimensional Dynamic Behavior Analysis of Leaves

To validate the effectiveness and reliability of the multi-dimensional dynamic behavior analysis method for leaves, this study conducted experiments on leaf fluttering patterns using the SAM2-FT. The dynamic characteristics of leaves were first collected, including time-series data on area, centroid position, and angle (sample data shown in Figure 12). The experimental dataset consisted of 10 groups of fluttering leaf videos processed with background segmentation.
The Hilbert transform was employed to extract the instantaneous frequency and amplitude envelope information of the leaf angle signals. Additionally, the instantaneous velocity of the leaf was calculated based on the motion trajectory of its centroid. Through the visualization of time-series curve plots and three-dimensional motion trajectory diagrams, the dynamic response characteristics of leaves under wind field conditions were analyzed.
The time-series analysis of instantaneous frequency, amplitude envelope, and angle (Figure 13a) provides precise quantification of leaf vibration patterns, revealing the fluctuation range of instantaneous frequency, the dynamic trends of the amplitude envelope, and the adaptive adjustment characteristics of inclination under wind field disturbances. This approach intuitively demonstrates the variations in vibration intensity and stability, offering robust data support for exploring the dynamic response mechanisms of leaves under complex wind field conditions.
Additionally, the combined analysis of instantaneous velocity and three-dimensional motion trajectories (Figure 13b) comprehensively uncovers the complex temporal and spatial characteristics of leaf dynamics. The temporal analysis of velocity curves reflects the dynamic changes in the intensity of leaf responses, while the visualization of three-dimensional motion trajectories reveals the overall movement patterns and spatial distribution of leaf trajectories under wind field influences. This lays a solid foundation for further investigation into the effects of wind field conditions on leaf motion behavior.
To analyze the dynamic characteristics of leaf area fluctuations during fluttering, amplitude and instantaneous rate were introduced, combined with the visualization of time-series curves and density distribution maps. This approach reveals the intensity, speed, and dynamic trends of leaf area variations, quantifying the sensitivity and adaptability of leaves in response to wind field disturbances.
The amplitude time-series analysis (Figure 14a) precisely quantifies the intensity and regularity of leaf area fluctuations, revealing the fluctuation range of area amplitudes, dynamic response characteristics, and the distribution patterns of area variations across different time intervals. This method intuitively demonstrates the dynamic features and stability differences of leaf area fluctuations.
Additionally, the time-area kernel density distribution (Figure 14b) systematically illustrates the concentration trends and time dependency of area variations, quantifying the dynamic characteristics of area changes in the temporal dimension. The distribution of area change rates (Figure 14c) further uncovers the intensity of fluctuations and rate characteristics, reflecting the range and patterns of area change rates. These findings provide scientific evidence for a deeper exploration of leaf dynamic behavior and stability.
The above experimental results indicate that the proposed multi-dimensional dynamic behavior analysis method for leaves demonstrates high effectiveness and reliability. This approach enables an in-depth analysis of the regularity of leaf fluttering characteristics from multiple dimensions, offering valuable insights for modeling leaf dynamic responses, improving spraying efficiency in precision agriculture, and optimizing wind field designs in agricultural environments.

4. Discussion

4.1. Performance Analysis of SAM2-FT

In the current research landscape, deep learning-based frameworks for object segmentation and tracking remain underdeveloped in the context of leaf motion detection under wind-driven conditions. To address this gap, this study proposes SAM2-FT, a Segment Anything Model-based framework incorporating cross-frame object segmentation and tracking, tailored for detecting wind-induced leaf vibrations. For comparison, three mainstream YOLO series models—YOLOv8x-seg, YOLOv9e-seg, and YOLOv10x-seg—were selected because of their strong real-time detection capabilities and widespread use in agricultural image analysis tasks. These models represent state-of-the-art spatial object detection approaches commonly applied in agricultural scenarios.
Experimental results indicate that SAM2-FT significantly outperforms the YOLO series models in terms of Precision, Recall, AP50, and F1-score. This performance advantage can be attributed to its integration of temporal sequence modeling, where inter-frame feature dependencies and attention mechanisms contribute to enhanced segmentation stability in complex dynamic environments. Unlike YOLO models, which rely solely on single-frame image features, SAM2-FT leverages spatiotemporal context across frames, enabling more accurate identification and tracking of leaf contours that exhibit continuous fluttering, rotation, and deformation in wind fields. Furthermore, its flexible, prompt mechanisms (e.g., point, box, or region-based prompts) and extensive pretraining on large-scale, diverse datasets enhance its generalization capacity across different crop types, imaging conditions, and complex backgrounds—particularly relevant to dynamic leaf detection tasks such as those explored in this study.
In the realm of wind-induced leaf vibration detection, existing mainstream approaches include Computational Fluid Dynamics (CFD) simulations and high-speed video analysis with point tracking. For instance, Qiu et al. utilized multiphase CFD models to simulate aerodynamic responses, successfully capturing dynamic leaf vibration behaviors and inclination angle changes [18]. Similarly, Cui et al. developed an aerodynamic response model for cotton leaves based on a combination of Finite Element (FE) and Lattice-Boltzmann (LB) methods, enabling the simulation of both aerodynamic responses and intra-canopy airflow distribution [27]. While CFD-based methods provide valuable insights into the fluid–structure interactions of leaves, their dependence on idealized boundary conditions and regularized leaf geometries limits their applicability in real-world, multi-leaf, and multi-scale dynamic environments. On the other hand, traditional video tracking methods often rely on single-point trajectory analysis (e.g., leaf tip or centroid), which fails to capture complex dynamic behaviors such as deformation, rotation, and fluttering of entire leaves. These methods also suffer from limited feature dimensions and strong dependence on manual annotation [34].
In summary, the segmentation and tracking framework proposed in this study—SAM2-FT—overcomes the theoretical constraints of CFD simulations and the dimensional limitations of single-point tracking approaches. It offers a high spatiotemporal resolution, non-invasive, full-frame solution for data acquisition and analysis of leaf dynamics, providing a novel and effective pathway for modeling wind-induced leaf behaviors.

4.2. Stability and Generalizability of the PCA-Based Background Separation Method

In this study, Principal Component Analysis (PCA) was employed for color space dimensionality reduction and clustering, achieving high-precision background separation in dynamic leaf videos, with the Intersection over Union (IoU) consistently exceeding 98.2% and exhibiting minimal error fluctuations. By maximizing the variance direction in these data, PCA performs linear dimensionality reduction, effectively extracting dominant variation features in high-dimensional spaces. Zhu et al. applied PCA to extract features from surface electromyography (sEMG) signals and further enhanced the algorithm by integrating kernel functions, enabling the modeling and extraction of nonlinear feature distributions in sEMG data [35]. Similarly, Yuan et al. utilized PCA to reduce the dimensionality of high-dimensional biological data and combined it with Bayesian variable selection to efficiently identify key variables in sparse latent factor structures. These studies highlight the capability of PCA to perform robust feature extraction in complex datasets [36].
Building upon this foundation, the PCA method adopted in this study facilitates the automatic selection of optimal segmentation thresholds between leaf regions and backgrounds. It enhances the adaptability and robustness of HSV-based thresholding by reducing reliance on fixed parameters. Compared with traditional fixed-threshold methods, the PCA-based approach minimizes manual subjectivity and significantly improves the algorithm’s adaptability to varying illumination and background conditions.
This method proves particularly effective in handling dynamic videos where the color pixel distribution of leaves varies substantially over time. By providing stable and adaptive background separation, it reduces noise in subsequent tracking and behavior analysis processes and mitigates cumulative inter-frame errors—thereby laying a solid foundation for accurate dynamic modeling in wind-induced leaf motion scenarios.
This study was conducted in a controlled indoor environment where the distinct color features of leaves allowed the HSV-based thresholding method to exhibit a certain degree of robustness. However, in unstructured orchard settings, increased background complexity may compromise segmentation performance. Future work will, therefore, focus on enhancing the adaptability of the approach under real-world orchard conditions.

4.3. Multi-Dimensional Dynamic Analysis for Quantifying Leaf Response Characteristics

Under wind-induced disturbances, leaves exhibit nonlinear dynamic behaviors such as rotation, bending, area fluctuation, and centroid drift. To capture these responses, this study employed the SAM2-FT model to accurately extract leaf angle, area, and centroid position across frames, constructing multi-dimensional temporal feature sequences. These sequences formed a comprehensive “morphology–posture–position” response representation framework.
For morphological response, inter-frame area variation sequences were used to calculate the Area Amplitude and the Rate of Area Change, which quantitatively characterize the leaf’s spatial expansion intensity and dynamic stability under wind excitation. For postural response, the time-varying inclination angles were analyzed using the Hilbert Transform, from which the envelope amplitude and instantaneous frequency were extracted to reflect the leaf’s posture adjustment strength and vibration rhythm, respectively. Regarding positional response, centroid trajectory tracking enabled the computation of instantaneous velocity within the image space, revealing the leaf’s motion direction, velocity fluctuations, and overall spatial dynamics. Together, these metrics enabled the construction of time- and frequency-domain response features that are both observable and quantifiable. This provided a high-resolution approach to capturing complex fluttering patterns, offering a robust data acquisition foundation for future wind–leaf interaction modeling.
In previous studies on quantifying leaf vibration responses, Li et al. developed a mechanical model under non-periodic excitation based on boundary layer theory and convolution integral methods. They derived an analytical expression for instantaneous leaf velocity and validated the model against empirical measurements [37]. Xu et al. proposed using leaf-tip trajectories and torsional angles to quantify motion amplitude and spatial posture variations, providing a non-contact approach to spatial dynamic measurements [38]. Compared with these approaches, the present study extends the analysis into a multi-indicator, multi-dimensional framework, establishing a systematic method for quantifying wind-driven leaf dynamics.
Unlike traditional CFD-based numerical simulations, which often rely on idealized boundary conditions, rigid structural assumptions, and predefined material properties, the method proposed in this study directly extracts time-series response features from real-world video data. By integrating deep learning-based segmentation with temporal modeling techniques, the framework achieves high-precision dynamic response extraction without the limitations of CFD in simulating large-scale, asymmetric deformations and rapid responses across multiple leaves. This approach thus offers a more scalable and realistic pathway for studying complex wind–leaf interaction phenomena in natural environments.

5. Conclusions

This study integrates an adaptively fine-tuned Segment Anything Model 2 (SAM2-FT) and multi-dimensional dynamic behavior analysis techniques to provide a systematic approach for uncovering the fluttering patterns of leaves under wind field disturbances. The main conclusions are as follows:
Innovation and reliability of the background segmentation algorithm: This study presents a PCA-based background segmentation algorithm, achieving high-precision segmentation in dynamic scenarios involving fluttering leaves. Experimental results demonstrate excellent performance in terms of Intersection over Union (IoU), Precision, Recall, and F1-score. In addition, the algorithm effectively extracts the key color features distinguishing the leaves from the background, enabling the determination of optimal HSV parameter ranges.
Efficient segmentation and tracking performance of SAM2-FT: Experimental results indicate that the SAM2-FT model achieved Precision, Recall, AP50, and F1 scores of 98.7%, 97.48%, 98.56%, and 98.08%, respectively, significantly outperforming YOLO series models such as YOLOv8x-seg and YOLOv10x-seg. By integrating time-series modeling with cross-frame consistency optimization, SAM2-FT enables high-precision segmentation and stable tracking in complex dynamic scenarios.
Multi-dimensional dynamic behavior analysis of leaves: This study established a multi-dimensional quantitative framework for analyzing leaf dynamic responses, integrating Hilbert transform and time-series analysis methods. The framework incorporates leaf area, inclination angle, and centroid displacement as core parameters. The results indicate that this approach enables a unified characterization of leaf vibration responses from three distinct dimensions: morphological features, postural variations, and spatial displacement information.
Academic significance and future outlook: This study presents a systematic methodology for the precise segmentation, tracking, and multi-dimensional dynamic response analysis of fluttering leaves under wind field conditions. To further advance this work, future research will emphasize the acquisition of time-series data and the development of data-driven predictive models to reveal the nonlinear influences of wind fields on leaf dynamics. In parallel, efforts will be devoted to optimizing detection algorithms and designing an efficient real-time environmental perception system. This system will be integrated into an autonomous spraying platform to enable real-time sensing and dynamic spraying control in mobile orchard environments. The implementation of this intelligent system is expected to significantly reduce pesticide usage and improve operational efficiency, thereby lowering agricultural production costs and providing robust technical support for the advancement of smart agriculture and precision farming.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/agriculture15080886/s1, Video S1: Pear Leaf Segmentation and Tracking.

Author Contributions

Conceptualization, Y.W. and S.D.; methodology, Y.W.; software, M.O.; validation, Y.W., W.J. and M.O.; formal analysis, Y.W.; investigation, B.G. and D.T.; resources, W.J.; data curation, X.D.; writing—original draft preparation, Y.W.; writing—review and editing, Y.W.; visualization, G.W.; supervision, W.J.; project administration, M.O.; funding acquisition, W.J. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (grant number: PAPD-2023-87), Research and Development of Mechanized Technology and Equipment for Key Stages in the Production of Wine Grapes and Daylilies, and Demonstration (8331203012) and the 2021 Provincial Key Laboratory—Research and Development of Key Technologies for Rice-Wheat Water, Fertilizer, and Pesticide Compound Operations and Integration of Intelligent Equipment (8261200003).

Institutional Review Board Statement

Not applicable.

Data Availability Statement

Data presented in this study are available on request from the corresponding author. These data are not publicly available due to confidentiality and privacy restrictions.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Liu, H.; Du, Z.; Shen, Y.; Du, W.; Zhang, X. Development and evaluation of an intelligent multivariable spraying robot for orchards and nurseries. Comput. Electron. Agric. 2024, 222, 109056. [Google Scholar] [CrossRef]
  2. Chen, C.; Li, S.; Wu, X.; Li, J.; Jia, Y.; Kang, F.; Wang, Y. Research on the deposition law of the spray droplet group based on single droplet multiphase flow simulation. J. Clean. Prod. 2023, 423, 138605. [Google Scholar] [CrossRef]
  3. Lin, J.; Ma, J.; Liu, K.; Huang, X.; Xiao, L.; Ahmed, S.; Dong, X.; Qiu, B. Development and test of an autonomous air-assisted sprayer based on single hanging track9 for solar greenhouse. Crop Prot. 2021, 142, 105502. [Google Scholar] [CrossRef]
  4. Dou, H.; Zhai, C.; Zhang, Y.; Chen, L.; Gu, C.; Yang, S. Research on decoupled air speed and air volume adjustment methods for air-assisted spraying in orchards. Front. Plant Sci. 2023, 14, 1250773. [Google Scholar] [CrossRef]
  5. Miranda, A.; Marucco, P.; González, E.J.; Gil, E.; Grella, M.; Balsari, P. Developing strategies to reduce spray drift in pneumatic spraying in vineyards: Assessment of the parameters affecting droplet size in pneumatic spraying. Sci. Total Environ. 2018, 616, 805–815. [Google Scholar] [CrossRef]
  6. Guo, J.; Dong, X.; Qiu, B. Analysis of the Factors Affecting the Deposition Coverage of Air-Assisted Electrostatic Spray on Tomato Leaves. Agronomy 2024, 14, 1108. [Google Scholar] [CrossRef]
  7. Shi, Q.; Liu, D.; Mao, H.; Shen, B.; Li, M. Wind-induced response of rice under the action of the downwash flow field of a multi-rotor UAV. Biosyst. Eng. 2021, 203, 60–69. [Google Scholar] [CrossRef]
  8. Cao, Y.; Wang, C.; An, Y.; Chen, Y.; Qiu, W. Droplet deposition behavior on the surface of wheat leaves with wind-induced vibration. Crop Protection 2024, 181, 106699. [Google Scholar] [CrossRef]
  9. Zhang, C.; Zhou, H.; Xu, L.; Ru, Y.; Ju, H.; Chen, Q. Wind tunnel study of the changes in drag and morphology of three fruit tree species during air-assisted spraying. Biosyst. Eng. 2022, 218, 153–162. [Google Scholar] [CrossRef]
  10. Gao, Z.M.; Hu, W.; Dong, X.Y.; Zhao, X.Y.; Wang, S.; Chen, J.; Qiu, B.J. Motion behavior of droplets on curved leaf surfaces driven by airflow. Front. Plant Sci. 2024, 15, 1450831. [Google Scholar] [CrossRef]
  11. Wu, S.; Liu, J.; Wang, J.; Hao, D.; Wang, R. The motion of strawberry leaves in an air-assisted spray field and its influence on droplet deposition. Trans. ASABE 2021, 64, 83–93. [Google Scholar] [CrossRef]
  12. Qin, W.C.; Chen, P.Y. Analysis of the research progress on the deposition and drift of spray droplets by plant protection UAVs. Sci. Rep. 2023, 13, 14935. [Google Scholar]
  13. Cui, H.; Wang, C.; Yu, S.; Xin, Z.; Liu, X.; Yuan, J. Two-stage CFD simulation of droplet deposition on deformed leaves of cotton canopy in air-assisted spraying. Comput. Electron. Agric. 2024, 224, 109228. [Google Scholar] [CrossRef]
  14. Wu, Z.; Liu, C.; Li, C.; Song, W.; Zhang, S. Establishment of fog droplet distribution model and study on canopy deposition uniformity. Phys. Fluids 2024, 36, 077113. [Google Scholar] [CrossRef]
  15. Jiang, S.; Yang, S.; Xu, J.; Li, W.; Zheng, Y.; Liu, X.; Tan, Y. Wind field and droplet coverage characteristics of air-assisted sprayer in mango-tree canopies. Pest Manag. Sci. 2022, 78, 4892–4904. [Google Scholar] [CrossRef] [PubMed]
  16. Qin, W.C.; Qiu, B.J.; Xue, X.Y.; Chen, C.; Xu, Z.F.; Zhou, Q.Q. Droplet deposition and control effect of insecticides sprayed with an unmanned aerial vehicle against plant hoppers. Crop Protection 2016, 85, 79–88. [Google Scholar] [CrossRef]
  17. Ma, J.; Liu, K.; Dong, X.; Huang, X.; Ahmad, F.; Qiu, B. Force and motion behaviour of crop leaves during spraying. Biosyst. Eng. 2023, 235, 83–99. [Google Scholar] [CrossRef]
  18. Qiu, W.; Guo, H.; Zheng, H.; Cao, Y.; Lv, X.; Fang, J.; Zhai CYu, H. CFD modelling to analyze the droplets deposition behavior on vibrating rice leaves. Comput. Electron. Agric. 2022, 201, 107330. [Google Scholar] [CrossRef]
  19. Xi, T.; Li, C.; Qiu, W.; Wang, H.; Lv, X.; Han, C.; Ahmad, F. Droplet deposition behavior on a pear leaf surface under wind-induced vibration. Appl. Eng. Agric. 2020, 36, 913–926. [Google Scholar] [CrossRef]
  20. Cao, Y.; Xi, T.; Xu, L.; Qiu, W.; Guo, H.; Lv, X.; Li, C. Computational fluid dynamics simulation experimental verification and analysis of droplets deposition behaviour on vibrating pear leaves. Plant Methods 2022, 18, 80. [Google Scholar] [CrossRef]
  21. Hu, Y.; Chen, Y.; Wei, W.; Hu, Z.; Li, P. Optimization Design of Spray Cooling Fan Based on CFD Simulation and Field Experiment for Horticultural Crops. Agriculture 2021, 11, 566. [Google Scholar] [CrossRef]
  22. Yan, C.; Niu, C.; Ma, S.; Tan, H.; Xu, L. CFD models as a tool to analyze the deformation behavior of grape leaves under an air-assisted sprayer. Comput. Electron. Agric. 2022, 198, 107112. [Google Scholar] [CrossRef]
  23. Xu, T.; Zhou, H.; Lv, X.; Lei, X.; Tao, S. Study of the distribution characteristics of the airflow field in tree canopies based on the CFD model. Agronomy 2022, 12, 3072. [Google Scholar] [CrossRef]
  24. Liu, Z.; Chen, J.; Guo, J.; Qiu, B. Numerical Simulation and Validation of Droplet Deposition on Tomato Leaf Surface under Air-Assisted Spraying. Agronomy 2024, 14, 1661. [Google Scholar] [CrossRef]
  25. Wu, S.; Liu, J.; Zhen, J.; Lei, X.; Chen, Y. Resistance characteristics of broad-leaf crop canopy in air-assisted spray field and their effects on droplet deposition. Front. Plant Sci. 2022, 13, 924749. [Google Scholar] [CrossRef] [PubMed]
  26. Cui, H.; Wang, C.; Liu, X.; Yuan, J.; Liu, Y. Dynamic simulation of fluid-structure interactions between leaves and airflow during air-assisted spraying: A case study of cotton. Comput. Electron. Agric. 2023, 209, 107817. [Google Scholar] [CrossRef]
  27. He, Y.; Wu, J.; Fu, H.; Sun, Z.; Fang, H.; Wang, W. Quantitative analysis of droplet size distribution in plant protection spray based on machine learning method. Water 2022, 14, 175. [Google Scholar] [CrossRef]
  28. Seol, J.; Kim, J.; Son, H.I. Field evaluations of a deep learning-based intelligent spraying robot with flow control for pear orchards. Precis. Agric. 2022, 23, 712–732. [Google Scholar] [CrossRef]
  29. Liu, J.; Abbas, I.; Noor, R.S. Development of Deep Learning-Based Variable Rate Agrochemical Spraying System for Targeted Weeds Control in Strawberry Crop. Agronomy 2021, 11, 1480. [Google Scholar] [CrossRef]
  30. Luo, Y.; Wei, L.; Xu, L.; Zhang, Q.; Liu, J.; Cai, Q.; Zhang, W. Stereo-vision-based multi-crop harvesting edge detection for precise automatic steering of combine harvester. Biosyst. Eng. 2022, 215, 115–128. [Google Scholar] [CrossRef]
  31. Zhao, X.; Guo, J.; Nie, F.; Chen, L.; Li, Z.; Zhang, H. Joint principal component and discriminant analysis for dimensionality reduction. IEEE Trans. Neural Netw. Learn. Syst. 2019, 31, 433–444. [Google Scholar] [CrossRef]
  32. Zhang, L.; Su, H.; Shen, J. Hyperspectral dimensionality reduction based on multiscale superpixelwise kernel principal component analysis. Remote Sens. 2019, 11, 1219. [Google Scholar] [CrossRef]
  33. Ma, C.; Zheng, L.; Li, X.; Wang, H.; Zhao, M.; Zhai, C. Comprehensive evaluation of power quality based on combination of minibatch K-means algorithm and random forest algorithm. In Proceedings of the Third International Conference on Artificial Intelligence and Electromechanical Automation, Changsha, China, 8–10 April 2022; Volume 12329, pp. 468–474. [Google Scholar]
  34. Zhang, C.; Zhou, H.; Xu, L.; Ru, Y.; Ju, H.; Chen, Q. Measurement of morphological changes of pear leaves in airflow based on high-speed photography. Front. Plant Sci. 2022, 13, 900427. [Google Scholar] [CrossRef]
  35. Zhu, M.; Guan, X.; Li, Z.; He, L.; Wang, Z.; Cai, K. sEMG-based lower limb motion prediction using CNN-LSTM with improved PCA optimization algorithm. J. Bionic Eng. 2023, 20, 612–627. [Google Scholar] [CrossRef]
  36. Yuan, D.; Mancuso, N. SuSiE PCA: A scalable Bayesian variable selection technique for principal component analysis. Iscience 2023, 26, 11. [Google Scholar] [CrossRef] [PubMed]
  37. Li, J.; Li, Z.; Ma, Y.; Cui, H.; Yang, Z.; Lu, H. Effects of leaf response velocity on spray deposition with an air-assisted orchard sprayer. Int. J. Agric. Biol. Eng. 2021, 14, 123–132. [Google Scholar] [CrossRef]
  38. Xu, L.; Wu, Y.; Zhou, H.; Zhang, J.; Zhang, C. Analysis of spatial motion attitude and droplet deposition effect of tree leaves in response to wind vibration. Trans. Chin. Soc. Agric. Eng. 2024, 40, 71–81. [Google Scholar]
Figure 1. Layout of trellis-based pear orchard and leaf fluttering data collection. In the laboratory-controlled environment, high-resolution and high-frame-rate leaf fluttering videos were captured using an action camera (4 K/120 fps) to record the dynamic behavior of leaves under wind fields precisely. The wind source was provided by an air compressor (300 L displacement, 7 bar rated pressure) to ensure wind field stability and adjustability, simulating realistic wind conditions. A universal clamp was employed to secure the leaves, ensuring their stability under controlled wind conditions during the experiment. This setup guarantees the reliability and repeatability of the data collection process.
Figure 1. Layout of trellis-based pear orchard and leaf fluttering data collection. In the laboratory-controlled environment, high-resolution and high-frame-rate leaf fluttering videos were captured using an action camera (4 K/120 fps) to record the dynamic behavior of leaves under wind fields precisely. The wind source was provided by an air compressor (300 L displacement, 7 bar rated pressure) to ensure wind field stability and adjustability, simulating realistic wind conditions. A universal clamp was employed to secure the leaves, ensuring their stability under controlled wind conditions during the experiment. This setup guarantees the reliability and repeatability of the data collection process.
Agriculture 15 00886 g001
Figure 2. PCA-based background separation of leaf images from video frames. The left panel shows a video frame image; the PCA 2D projection visualizes the distribution of dimensionality-reduced HSV color information, highlighting the separation characteristics between the background and the leaf. The Hue-Saturation and Hue-Value distribution plots reflect the clustered features and specific thresholds of these reduced HSV data, enabling precise separation of the leaf from the background.
Figure 2. PCA-based background separation of leaf images from video frames. The left panel shows a video frame image; the PCA 2D projection visualizes the distribution of dimensionality-reduced HSV color information, highlighting the separation characteristics between the background and the leaf. The Hue-Saturation and Hue-Value distribution plots reflect the clustered features and specific thresholds of these reduced HSV data, enabling precise separation of the leaf from the background.
Agriculture 15 00886 g002
Figure 3. Mask images of fluttering Leaves. The images in (ac) are fluttering leaf mask images accurately annotated using the LabelMe tool.
Figure 3. Mask images of fluttering Leaves. The images in (ac) are fluttering leaf mask images accurately annotated using the LabelMe tool.
Agriculture 15 00886 g003
Figure 4. Schematic of the SAM2 framework for dynamic leaf segmentation and tracking.
Figure 4. Schematic of the SAM2 framework for dynamic leaf segmentation and tracking.
Agriculture 15 00886 g004
Figure 5. Schematic of leaf fluttering characteristic variables: area, centroid, and inclination angle. (a) Area: the unit is square pixels (px2), representing the projected area of the leaf in the image space; (b) Angle: the unit is degrees (°), representing the angle of the leaf’s inclination relative to the horizontal baseline; (c) Centroid: the coordinate unit is pixels (px), relative to the top-left corner of the image.
Figure 5. Schematic of leaf fluttering characteristic variables: area, centroid, and inclination angle. (a) Area: the unit is square pixels (px2), representing the projected area of the leaf in the image space; (b) Angle: the unit is degrees (°), representing the angle of the leaf’s inclination relative to the horizontal baseline; (c) Centroid: the coordinate unit is pixels (px), relative to the top-left corner of the image.
Agriculture 15 00886 g005
Figure 6. (ad) Schematic of sample images for the segmentation experiment.
Figure 6. (ad) Schematic of sample images for the segmentation experiment.
Agriculture 15 00886 g006
Figure 7. (ad) Segmentation results of fluttering leaves using SAM2-FT.
Figure 7. (ad) Segmentation results of fluttering leaves using SAM2-FT.
Agriculture 15 00886 g007
Figure 8. (ad) Sample images of video frames showing leaf fluttering.
Figure 8. (ad) Sample images of video frames showing leaf fluttering.
Agriculture 15 00886 g008
Figure 9. (ad) PCA Analysis of Leaf Color Information. The PCA 2D projection illustrates the distribution of dimensionality-reduced HSV color information. The Hue-Saturation and Hue-Value distribution plots depict the clustered HSV features and specific thresholds derived from these reduced data, which are used for the precise separation of leaves from the background.
Figure 9. (ad) PCA Analysis of Leaf Color Information. The PCA 2D projection illustrates the distribution of dimensionality-reduced HSV color information. The Hue-Saturation and Hue-Value distribution plots depict the clustered HSV features and specific thresholds derived from these reduced data, which are used for the precise separation of leaves from the background.
Agriculture 15 00886 g009
Figure 10. Heatmap of HSV Threshold Distribution for Leaves.
Figure 10. Heatmap of HSV Threshold Distribution for Leaves.
Agriculture 15 00886 g010
Figure 11. (ad) Schematic of background segmentation for fluttering leaves.
Figure 11. (ad) Schematic of background segmentation for fluttering leaves.
Agriculture 15 00886 g011
Figure 12. Leaf fluttering characteristic variables: Area, Angle, and Centroid.
Figure 12. Leaf fluttering characteristic variables: Area, Angle, and Centroid.
Agriculture 15 00886 g012
Figure 13. Time-series analysis of leaf angle and centroid dynamics. (a) Leaf Angle represents the time-series variation of the leaf angle, Amplitude Envelope depicts the time-series changes in vibration intensity (amplitude), and Instantaneous Frequency shows the time-dependent variation in the instantaneous frequency of leaf vibrations; (b) Leaf Speed illustrates the time-series variation of leaf motion speed, while 3D Motion Trajectory represents the leaf’s movement trajectory in three-dimensional space.
Figure 13. Time-series analysis of leaf angle and centroid dynamics. (a) Leaf Angle represents the time-series variation of the leaf angle, Amplitude Envelope depicts the time-series changes in vibration intensity (amplitude), and Instantaneous Frequency shows the time-dependent variation in the instantaneous frequency of leaf vibrations; (b) Leaf Speed illustrates the time-series variation of leaf motion speed, while 3D Motion Trajectory represents the leaf’s movement trajectory in three-dimensional space.
Agriculture 15 00886 g013
Figure 14. Multi-view analysis of dynamic characteristics of leaf area fluctuations. (a) Time-series of leaf area vibrations; (b) Joint density distribution of time and area variations; (c) Probability distribution of the rate of leaf area change.
Figure 14. Multi-view analysis of dynamic characteristics of leaf area fluctuations. (a) Time-series of leaf area vibrations; (b) Joint density distribution of time and area variations; (c) Probability distribution of the rate of leaf area change.
Agriculture 15 00886 g014
Table 1. Results of model comparison experiments.
Table 1. Results of model comparison experiments.
ModelPrecision/%Recall/%AP50/%F1
YOLOv9e-seg88.5085.1591.3086.79
YOLOv8x-seg88.8183.6290.2086.31
YOLOv10x-seg89.5685.1491.1586.98
SAM2-FT98.797.4898.5698.08
Table 2. Evaluation of background segmentation results for fluttering leaves.
Table 2. Evaluation of background segmentation results for fluttering leaves.
GroupsIoU/%Precision/%Recall/%F1
leafV_Seg198.599.398.298.7
leafV_Seg298.499.298.198.6
leafV_Seg398.299.098.098.5
leafV_Seg498.699.498.498.9
leafV_Seg598.799.598.699.0
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

Wang, Y.; Jia, W.; Dai, S.; Ou, M.; Dong, X.; Wang, G.; Gao, B.; Tu, D. Analytical Methods for Wind-Driven Dynamic Behavior of Pear Leaves (Pyrus pyrifolia). Agriculture 2025, 15, 886. https://doi.org/10.3390/agriculture15080886

AMA Style

Wang Y, Jia W, Dai S, Ou M, Dong X, Wang G, Gao B, Tu D. Analytical Methods for Wind-Driven Dynamic Behavior of Pear Leaves (Pyrus pyrifolia). Agriculture. 2025; 15(8):886. https://doi.org/10.3390/agriculture15080886

Chicago/Turabian Style

Wang, Yunfei, Weidong Jia, Shiqun Dai, Mingxiong Ou, Xiang Dong, Guanqun Wang, Bohao Gao, and Dengjun Tu. 2025. "Analytical Methods for Wind-Driven Dynamic Behavior of Pear Leaves (Pyrus pyrifolia)" Agriculture 15, no. 8: 886. https://doi.org/10.3390/agriculture15080886

APA Style

Wang, Y., Jia, W., Dai, S., Ou, M., Dong, X., Wang, G., Gao, B., & Tu, D. (2025). Analytical Methods for Wind-Driven Dynamic Behavior of Pear Leaves (Pyrus pyrifolia). Agriculture, 15(8), 886. https://doi.org/10.3390/agriculture15080886

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop