Peanut plant type low-altitude remote sensing monitoring method and system driven by canopy structure perception

Abstract

The present application belongs to the technical field of low-altitude unmanned aerial vehicle remote sensing and precision agriculture, and discloses a peanut plant type low-altitude remote sensing monitoring method and system driven by crown layer structure perception, which obtains high-resolution images through a low-altitude unmanned aerial vehicle, generates a digital surface model by combining POS alignment and image reconstruction, calculates a crown layer height model and estimates the crown layer height; removes weed interference dynamically through DBSCAN clustering and estimates the crown layer width; explores the feasibility analysis of driving peanut plant type recognition by using the width-height ratio parameter obtained by crown layer structure perception, and realizes automatic peanut plant type recognition based on the crown layer width-height ratio by using a decision tree classification model; finally, the dynamic change curves of the crown layer height and width are fitted by combining the Richards growth model, the peanut crown layer structure parameters in a long time sequence in a complete growth period are simulated, the best growth period window for realizing peanut plant type recognition by using the crown layer width-height ratio is determined by performing ANOVA analysis on each growth period window.

Description

Technical Field

[0001] This invention belongs to the field of low-altitude unmanned aerial vehicle remote sensing and precision agriculture technology, and discloses a method and system for low-altitude remote sensing monitoring of peanut plant type driven by canopy structure perception. Background Technology

[0002] As a typical underground seed-bearing crop, peanuts' above-ground canopy structure largely determines flower bud development, pod formation, and field ventilation. Crop phenotyping typically involves analyzing canopy structure and morphological characteristics, such as plant type, canopy height, canopy width, and canopy coverage. Among these, plant type, as a comprehensive indicator of crop structural phenotyping, is a crucial morphological indicator characterizing individual growth vigor and canopy light utilization efficiency. It directly impacts light interception, pest and disease ventilation, and field management efficiency, and has become a key parameter in agronomic research, breeding screening, and field management. Therefore, accurate identification of crop plant type is essential for crop phenotypic trait analysis and precise identification of germplasm resources.

[0003] Traditional crop plant architecture assessment relies primarily on manual field observation, which is inefficient, labor-intensive, and unsuitable for large-scale, rapid data acquisition. With the rapid development of UAV remote sensing technology, high-resolution imagery provides a feasible approach for non-contact, high-throughput monitoring of crop canopy structure, and numerous studies have been conducted on extracting crop phenotypic data using UAV remote sensing technology. Peanut plant architecture is typically characterized by canopy structure elements such as main stem height, first branch length, number of branches, and branch angle, and can be classified into four basic types: creeping, semi-creeping, semi-upright, and upright. Accurate perception and extraction of peanut canopy structure data is the first and most crucial step in peanut plant architecture identification.

[0004] In agronomy, although there are various definitions of peanut plant architecture based on canopy structure, rapidly obtaining plant architecture information at different growth stages remains challenging. This is mainly because most plant architecture classification indicators (such as branch angles and growth patterns) rely on field measurements, making them difficult to obtain. Since peanuts bear fruit underground, the distance between the plant and the ground becomes a crucial factor influencing pod formation. The maximum main stem height (MSH) and the longest lateral branch length (LBL) are considered key parameters characterizing peanut plant architecture. In summary, MSH and LBL play important roles in the characteristics of peanut plant architecture. Therefore, from the perspective of canopy structure perception, peanut plant architecture can be considered to be driven by two main canopy structures: canopy height and canopy width. To improve recognition efficiency and automation, this invention proposes a structural feature expression framework based on the ratio of canopy width to height (width-to-height ratio) to achieve rapid identification of different peanut plant architectures. Summary of the Invention

[0005] To overcome the problems existing in related technologies, this invention provides a low-altitude remote sensing monitoring method for peanut plant architecture driven by canopy structure perception, comprising:

[0006] S1. Based on high-resolution RGB images, complete POS alignment and image reconstruction, generate digital surface model (DSM) and digital elevation model (DEM), and construct canopy height model (CHM) to estimate canopy height;

[0007] S2. Quantitatively extract the visual features of peanut canopy and soil background, dynamically remove weed interference by DBSCAN clustering and estimate canopy width;

[0008] S3. Based on the aspect ratio features extracted from canopy structure perception, construct a decision tree classification model based on aspect ratio to achieve automatic identification of different peanut plant types;

[0009] S4. Based on the growth fitting curve, time-series modeling of peanut canopy height and width is performed to obtain the dynamic change characteristics of canopy width-to-height ratio during the growth period. Key growth periods with plant type discrimination ability are identified from the time series, realizing accurate identification of peanut plant type based on canopy structure perception.

[0010] Furthermore, in step S1, the digital surface model (DSM) of the i-th peanut sub-region is used. i With Digital Elevation Model (DEM) i CHM model for calculating the canopy height of the i-th peanut subplot using pixel-by-pixel interpolation. i This enables precise perception of canopy height structure. The calculation formula is as follows:

[0011] CHM i =DSM i -DEM i ;

[0012] Where i represents the corresponding peanut sub-region plot.

[0013] Then, based on CHM i The peanut canopy height for each plot was extracted using percentile statistics.

[0014] Furthermore, in step S2, effective separation of the peanut canopy and soil is performed based on the color index EXG. Specifically, the three color channels of the RGB image are first normalized, as shown in the expression:

[0015] EXG = 2g-rb;

[0016] in,

[0017]

[0018] In the formula, R, G, and B are the channel values ​​of the red, green, and blue channels, respectively.

[0019] Furthermore, based on the EXG index image, a fixed threshold method was used to binarize the image to obtain the peanut canopy region mask image (M) at each growth stage. t The expression is:

[0020]

[0021] In the formula, t represents the t-th reproductive period, and [i,j] represents the pixel position. For example, pixel [i,j] is identified as the peanut canopy region during the t-th growth stage. Pixel [i,j] is identified as soil during the t-th growth period.

[0022] Furthermore, the canopy masking maps (M) at each reproductive stage were analyzed. t Perform pixel-level logical AND operations to generate a cumulative temporal mask (M), expressed as:

[0023]

[0024] In the formula, M is the cumulative phase mask, T is the sum of all reproductive periods, and M t For the binary mask image generated during the t-th reproductive period, Let pixel (i,j) be identified as a vegetation area during the t-th growth stage. Let (i,j) be the pixel that was not identified as a vegetation area during the t-th growth period, where i and j are the row and column of the pixel's location.

[0025] Furthermore, the DBSCAN clustering algorithm is used to cluster connected canopy pixels, and the extracted single-temporal peanut canopy mask image (M) is then used to... t The data is divided into different pixel clusters. Based on the cumulative temporal mask (M), it is determined whether each pixel cluster exists continuously throughout the entire growth period of peanuts. If the current pixel cluster intersects with the cumulative temporal mask, then the pixel cluster is a peanut canopy region. Otherwise, the pixel cluster is determined to be a weed region, and weeds are removed. This process is repeated to mark and remove weeds throughout the entire growth period of peanuts.

[0026] Further, in step S3, a one-way analysis of variance (ANOVA) is used to test the significance of the peanut canopy width-to-height ratio in plant type identification, and the F-value (F-value) is used to measure the relative strength of differences between groups and within groups. _value )for:

[0027]

[0028] In the formula:

[0029]

[0030] Where MSB is the between-group mean square, MSW is the within-group mean square, and n i Let be the number of samples in the i-th group. Let be the mean of the i-th group. Y represents the population mean, k is the number of categories; N is the total number of samples. ij This represents the j-th observation in the i-th group.

[0031] Furthermore, when the means of different groups are exactly the same, the p-value of the current observation is obtained (p0). _value )for:

[0032] p _value =P(F (k-1,N-k) ≥F _value );

[0033] In the formula, F (k-1,N-k) It is an F-distribution with k-1 and Nk degrees of freedom;

[0034] Using a significance level of 0.05 as the judgment criterion, p _value If the aspect ratio is less than 0.05, it proves that the peanut plant type can be determined by the aspect ratio.

[0035] Furthermore, in step S3, the DT algorithm is used to complete the multi-class classification task based on aspect ratio. By recursively dividing the feature space, minimizing the Gini coefficient is used as the objective function to generate classification nodes layer by layer, thereby realizing the automatic identification of different peanut plant types.

[0036]

[0037] In the formula, p k This represents the proportion of the k-th class of samples in the current node.

[0038] Furthermore, in step S4, the Richards growth fitting model is introduced to perform time-series fitting on key structural indicators of canopy height and width.

[0039]

[0040] In the formula, y(t) represents the growth at time t, S is the maximum value of the model, r is the growth rate, A is a constant related to the initial conditions, and v is the shape parameter.

[0041] This invention also proposes a low-altitude remote sensing monitoring system for peanut plant type driven by canopy structure perception, comprising: a peanut canopy height structure perception and processing module, a peanut canopy width structure perception and processing module, a peanut plant type identification module, and a peanut plant type accurate temporal identification and window determination module.

[0042] The peanut canopy height structure perception and processing module completes POS alignment and image reconstruction based on high-resolution RGB images, generates a digital surface model (DSM) and a digital elevation model (DEM), and constructs a canopy height model (CHM).

[0043] The peanut canopy width structure perception and processing module quantitatively extracts the visual features of the peanut canopy and soil background, dynamically removes weed interference and estimates the canopy width through DBSCAN clustering;

[0044] The peanut plant type identification module constructs a decision tree classification model based on the aspect ratio features extracted from canopy structure perception to achieve automatic identification of different peanut plant types.

[0045] The peanut plant type accurate temporal identification and window determination module performs temporal modeling of peanut canopy height and canopy width based on growth fitting curves, obtains the dynamic change characteristics of canopy width-to-height ratio during the growth period, identifies key growth periods with plant type discrimination ability from the time series, and realizes accurate identification of peanut plant type based on canopy structure perception.

[0046] The beneficial effects of this invention are as follows:

[0047] This invention utilizes high-resolution RGB images acquired by UAV low-altitude remote sensing technology to complete the perception and extraction of peanut canopy structure, making the identification of peanut plant type based on canopy structure perception and aspect ratio driven simple and easy within a specified growth period.

[0048] This invention utilizes the high-efficiency, low-cost, high spatial resolution image acquisition capabilities of low-altitude UAV remote sensing, combined with the advantages of computer vision in canopy structure perception and processing, to generate a DSM (Digital Smart Scale) comparable to lidar point clouds by 3D reconstruction of small peanut plots, thereby improving the ease and accuracy of peanut canopy height estimation.

[0049] This invention achieves temporal information fusion through time-series images of peanut canopy masks, significantly improving the robustness of weed removal in peanut fields and enabling reliable separation of planting soil, thereby enhancing the accuracy and quality of crop canopy width estimation. Based on computer-aided ANOVA analysis and a decision tree machine learning model, it provides a complete technical path and evaluation system for plant type identification based on canopy structure perception-derived aspect ratio, from feasibility verification to assessment of indicators such as classification accuracy and recall.

[0050] This invention, based on the Richards growth model, provides a complete process for peanut plant architecture methods driven by canopy structure perception and aspect ratio, from structural time-series fitting to feature discriminative power evaluation, enabling precise determination of key growth period identification windows in plant architecture classification. This invention's canopy structure perception-driven low-altitude remote sensing monitoring technology for peanut plant architecture improves the automation and intelligence level of peanut plant architecture identification, providing a fundamental technical means for precision management, high-throughput phenotypic monitoring, and digital breeding in peanut agriculture. Attached Figure Description

[0051] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0052] Figure 1 This is a flowchart of the low-altitude remote sensing monitoring method for peanut plant type driven by canopy structure perception provided in an embodiment of the present invention;

[0053] Figure 2 This is a peanut plant type classification diagram provided in an embodiment of the present invention;

[0054] Figure 3 This is a diagram showing the feasibility results of ANOVA analysis based on canopy structure perception-width-to-height ratio driven peanut plant type identification provided in this embodiment of the invention;

[0055] Figure 4 This is a Richards model fitting diagram of the canopy structure of different peanut plant types provided in the embodiments of the present invention;

[0056] Figure 5 This is a diagram showing the ANOVA analysis results of the daily aspect ratio based on Richards fitting, provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0058] like Figure 1 As shown, the canopy structure sensing-driven low-altitude remote sensing monitoring method for peanut plant architecture provided in this embodiment of the invention includes the following steps:

[0059] S1. Based on high-resolution RGB images, POS alignment and image reconstruction are completed to generate a digital surface model (DSM) and a digital elevation model (DEM), and a canopy height model (CHM) is constructed to achieve accurate perception of peanut canopy height structure and canopy height estimation.

[0060] In a specific embodiment, peanut is used as the research object. High-resolution RGB image sequences are acquired using low-altitude UAV remote sensing technology. Canopy regions at different growth stages of peanut are extracted, and a crop canopy structure perception and recognition method based on UAV remote sensing is designed. Specifically, it includes:

[0061] This invention utilizes drone imagery data collected at different growth stages of peanuts. The peanut drone imagery data was collected and compiled in a peanut experimental field in a certain county, and the labels were manually added using labelme. This county has a typical warm temperate semi-humid continental climate with four distinct seasons, simultaneous rainfall and heat, and an average annual temperature of approximately 14℃. The county has an average altitude of 85 meters, an average annual sunshine duration of 2759.1 hours, an average annual precipitation of 901.4 mm, and an average frost-free period of 199 days, making it suitable for peanut growth. Data collection began on May 22, 2024, and ended on September 22, 2024, covering the peanut emergence, seedling, flowering and pegging, and pod-setting stages. The experiment used a DJI Mavic 3 drone equipped with a visible light camera, collecting data at midday on a clear, cloudless day. The DJI GS Pro platform was used for flight path planning and display, with the lateral overlap set to 80% and the flight altitude set to 12 meters. In DJI Terra software, flight strip stitching and radiometric correction are performed to generate digital orthophotos. Then, ArcMap 10.4 software is used to complete image registration and geographic coordinate system matching for peanut images from different periods, and shapefiles for each peanut plot are drawn.

[0062] In this invention, based on field observations and the morphological characteristics of peanut plants, peanut plant types are further subdivided into five types: upright, semi-upright, semi-prostrate, spreading, and prostrate. This plant type classification standard is not only based on previous research but also incorporates morphological parameters such as the angle between lateral branches and the main stem, and growth posture. Figure 2 As shown.

[0063] To accurately obtain the vertical structure information of peanut plants, traditional methods based on image segmentation or single height projection are insufficient to reflect the true height of the crop relative to the terrain and are easily affected by terrain undulations, image distortion, and noise interference. Therefore, this invention employs a 3D point cloud generation strategy based on the fusion of high-resolution RGB imagery and DSM imagery to recover the surface spatial structure information, including the crop canopy, and generate a DSM with centimeter-level spatial resolution. The DSM layer acquired during the bare soil period is defined as the DEM and used to eliminate the interference of vegetation on terrain elevation estimation. The CHM is obtained by calculating the pixel-by-pixel difference between the DSM and DEM during the growth period, achieving a true perception of the peanut canopy height. The process is as follows: Figure 2 As shown, it specifically includes:

[0064] Step 101, 3D reconstruction of the image.

[0065] During DSM generation, spatial reference information is extracted from the DSM and combined with pixel resolution to calculate the geographic coordinates (X, Y) of each pixel. The DSM value is then assigned as the Z-coordinate to construct an initial 3D geometric point cloud. Subsequently, each 3D point is mapped to the pixel coordinate system of the RGB image using coordinate projection, extracting the red, green, and blue (R, G, B) channel values ​​of the corresponding pixel to achieve effective fusion of point cloud geometry and color information. To ensure the legality and accuracy of the projection process, a boundary detection mechanism is introduced to correct row and column indices exceeding the image range, preventing data access errors caused by index out-of-bounds errors. Furthermore, to improve the quality and stability of the point cloud data, the program performs validity screening on elevation data during processing, removing all abnormal pixels with elevation values ​​less than a set threshold (-1000) to prevent background areas or erroneous pixels from being written into the point cloud file. Finally, all points with valid elevation values ​​are retained, forming xyzRGB point cloud data containing color attributes.

[0066] Step 102, CHM i Obtain.

[0067] Based on DEM i With DSM i The obtained CHM i This accurately reflects the true vertical structure of peanut plants relative to the actual terrain in a small area. The calculation formula is:

[0068] CHM i =DSM i -DEM i ;

[0069] Where i represents the corresponding peanut sub-region plot.

[0070] Based on CHM iThe peanut canopy height for each plot was extracted using percentile statistics.

[0071] Step 2: Quantitatively extract the visual features of the peanut canopy and soil background, dynamically remove weed interference through DBSCAN clustering, and estimate the canopy width.

[0072] Color features are among the most fundamental and robust visual features in image processing, widely used in computer vision and crop image analysis. Compared to features such as shape and texture, color features are less dependent on image size, orientation, and viewpoint, making them suitable for target recognition tasks in complex field scenes. Addressing the problem of severe weed interference before peanut seedling canopy closure, and the high visual similarity between weeds and peanuts, this study, based on the agronomic logic that "weeds do not persist throughout all growth stages," multiplies multi-temporal masks pixel-by-pixel to generate a stable canopy accumulation mask, which serves as the criterion for weed removal. Furthermore, the density-based clustering algorithm DBSCAN is applied to each single-temporal canopy mask to identify and segment pixel clusters. DBSCAN, as a density-based clustering algorithm, has significant advantages in spatial target recognition and image segmentation. Its core is based on setting the neighborhood radius and minimum sample number, analyzing the spatial distance relationships between points to achieve automatic clustering of high-density areas and identification of low-density noise points. By repeating this process on all time-phase maps, dynamic identification and full-growth-cycle removal of weed areas can be achieved, significantly improving the purity of canopy structure extraction and the accuracy of subsequent analysis. The process is as follows: Figure 2 As shown, specifically:

[0073] Step 201: Use the Supergreen Index (ExG) to separate the visual feature patterns of the peanut canopy from the soil background. The Supergreen Index (ExG) is used to calculate the greenness of the crop, and the formula is as follows:

[0074] ExG = 2g-rb;

[0075] In the formula, ExG is the super green index, g is the green band of the image, r is the red band of the image, and b is the blue band of the image.

[0076] The formulas for calculating r, g, and b are:

[0077]

[0078] In the formula, R, G, and B are the channel values ​​of the red, green, and blue channels, respectively;

[0079] Step 202, using a dynamic weed removal method based on RGB image time series, includes:

[0080] Based on the EXG index image, a fixed threshold method was used to binarize the image to obtain the peanut canopy region mask image (M) at each growth stage.t The expression is:

[0081]

[0082] In the formula, t represents the t-th reproductive period, and [i,j] represents the pixel position. For example, pixel [i,j] is identified as the peanut canopy region during the t-th growth stage. Pixel [i,j] is identified as soil during the t-th growth period.

[0083] Canopy masking at each reproductive stage (M) t Perform pixel-level logical AND operations to generate a cumulative temporal mask (M), expressed as:

[0084]

[0085] In the formula, M is the cumulative phase mask, T is the sum of time points, and M t For the binary mask image generated during the t-th reproductive period, Let pixel [i,j] be identified as a vegetation area at time t. Let [i,j] be the pixel that was not identified as a vegetation area during the t-th growth period, where i and j are the row and column of the pixel's location.

[0086] Step 203, combining the cumulative temporal mask image, performs DBSCAN cluster analysis to remove weeds, including:

[0087] The DBSCAN clustering algorithm is used to cluster connected canopy pixels, and the extracted single-temporal peanut canopy mask image M is obtained. t The data is divided into different pixel clusters. Based on the cumulative temporal mask M, it is determined whether each pixel cluster persists throughout the entire growth period of the peanut. If the current pixel cluster intersects with the cumulative temporal mask, then the pixel cluster is a peanut canopy region. Otherwise, the pixel cluster is determined to be a weed region, and the weeds are removed. This process is repeated to mark and remove weeds throughout the entire growth period of the peanut.

[0088] Step 204: Based on DBSCAN clustering analysis, the peanut canopy width is extracted. Morphological erosion is performed on the peanut image using a 3×3 structuring element to remove isolated noise and edge defects. Aperture filling and Canny edge detection are then used to enhance the connectivity of the canopy boundaries. Subsequently, DBSCAN clustering is used to identify effective canopy clusters, and each cluster is further subdivided along the row direction. The maximum Euclidean distance between pixels is calculated as the local canopy width. Based on this, a percentile statistical method is used to extract representative width values ​​as key parameters for perceiving the lateral structure of the canopy.

[0089] Step 3: Based on the aspect ratio features extracted from canopy structure perception, construct a decision tree classification model based on aspect ratio to achieve automatic identification of different peanut plant types.

[0090] First, one-way analysis of variance (ANOVA), a widely used feature discrimination technique in statistics, has the advantages of strong interpretability and high computational efficiency. In plant type identification and classification based on peanut canopy structure perception, ANOVA can be used to verify the significance of differences in a certain structural indicator between different categories, thereby determining whether it has discriminative ability. Compared with empirical methods or subjective judgment, ANOVA provides an objective and quantitative standard for feature selection, and is particularly suitable for feature effectiveness screening in the exploratory modeling stage. In this invention, ANOVA is used to test the distinguishing ability of peanut canopy width-to-height ratio between different plant types, providing a scientific basis for constructing subsequent classification models.

[0091] Subsequently, decision tree models, due to their intuitive structure, clear classification logic, and strong adaptability to nonlinear relationships and missing values, have been widely used in agricultural remote sensing and crop phenotypic identification. Decision trees can directly classify canopy structure ratio indicators (width-to-height ratio) without complex feature transformations, exhibiting good interpretability and practicality. This invention uses the width-to-height ratio as an input feature to construct a peanut plant type identification model based on measured data. The accuracy and stability of the model in multi-class identification tasks are evaluated through various performance evaluation indicators to ensure the reliability of the classification results.

[0092] Step 301: Using previously employed remote sensing and computer vision techniques, the aspect ratio parameters of the peanut canopy structure are extracted from the image, and plant type classification and identification are performed accordingly. The formula for the vegetation aspect ratio is as follows:

[0093]

[0094] Step 302: ANOVA analysis was used to statistically test the discriminative ability of peanut canopy width-to-height ratio in plant type identification. By constructing variance comparison models between and within different plant type groups, the F-value (F...) was analyzed. -value The p-value is used to measure the relative strength of differences between groups and within groups, while the p-value (p0.05) is used to measure the relative strength of differences between groups and within groups. -value The statistical significance of the difference is assessed based on the statistical distribution corresponding to the F-value, which is a key indicator in ANOVA analysis. The results of an example are as follows: Figure 3 As shown. The formula for calculating the F-value in ANOVA is:

[0095]

[0096] In the formula, Mean Square Between (MSB) measures the variability between the means of different classes, and is calculated as follows:

[0097]

[0098] Where n i Let be the number of samples in the i-th group. Let be the mean of the i-th group. K represents the overall mean, and K represents the number of categories.

[0099] In the formula, Mean Square Within (MSW) measures the data variability within each category, and is calculated as follows:

[0100]

[0101] Where N is the total number of samples, Y ij This typically represents the j-th observation in the i-th group. A large F-value indicates that the means of different categories differ significantly, meaning that phenotypic features such as aspect ratio can be used to distinguish between categories.

[0102] In ANOVA, the p-value is a key indicator in hypothesis testing. Specifically, the p-value calculated by ANOVA is used to test whether there are significant differences between the mean width-to-height ratios of different plant types. The null hypothesis (H0) is that the mean width-to-height ratios of different plant types are the same; the alternative hypothesis (H1) is that at least one group of types has a mean width-to-height ratio that is significantly different from the other groups. The p-value corresponding to the F-statistic obtained from ANOVA analysis is defined as the probability of obtaining the current observation result when the means of different groups are exactly the same. The formula for calculating the p-value is:

[0103] p _value =P(F (k-1,N-k) ≥F _value )

[0104] In the formula F (k-1,N-k) It is an F-distribution with k-1 and Nk degrees of freedom; k is the number of peanut plant types; N is the total number of samples; P represents the probability of probability; p _value The p-value is the probability, in the F-distribution, of a value to the right (or more extreme) that is larger than the current F-value; it is the definition of the p-value. ANOVA uses the p-value to test hypotheses, typically with a significance level of 0.05 as the criterion. _value If the mean is less than 0.05, the null hypothesis is rejected, indicating that at least one set of means is significantly different from the other sets, proving that the aspect ratio can be used to distinguish peanut plant types.

[0105] Step 303: Based on measured plant type data, a peanut plant type identification model based on canopy aspect ratio characteristics is constructed, and the DT algorithm is used to complete the multi-class classification task. This model recursively partitions the feature space, using minimizing the Gini coefficient as the objective function, and generates classification nodes layer by layer to achieve automatic identification of different peanut plant types. The standard formula for decision tree partitioning is:

[0106]

[0107] In the formula, p k This represents the proportion of the k-th class of samples in the current node.

[0108] Based on this, the following performance evaluation metrics were further calculated, including macro-average accuracy, weighted precision, weighted recall, and weighted F1 score. 1-score Multiple indicators were used to analyze the stability and distinguishability of the model in different plant type categories. The results of the implementation examples are shown in Table 1.

[0109] Table 1. Indicator Table Based on Decision Tree

[0110] Days after sowing Macro Accuracy Weighted Precision Weighted Recall <![CDATA[Weighted F 1-score ]]> 20 0.96 0.96 0.96 0.96 25 0.96 0.96 0.96 0.96 34 0.95 0.96 0.95 0.95 42 0.97 0.97 0.97 0.97 49 0.98 0.98 0.98 0.98 54 0.98 0.98 0.98 0.98 63 0.96 0.96 0.96 0.96 70 0.96 0.96 0.96 0.96

[0111] Macro-average accuracy, as a crucial metric in decision tree model analysis, measures the consistency of the model's overall classification performance across different plant types. This metric, by calculating the classification accuracy for each category separately and then taking the arithmetic mean, avoids bias caused by uneven category distribution in the overall evaluation results, directly reflecting the model's balanced discrimination performance across categories in multi-class recognition tasks. Accuracy helps the system verify the universality and effectiveness of the canopy structure-based feature system in peanut plant type recognition, serving as an important reference for comprehensive model performance evaluation and subsequent optimization. The formula for macro-accuracy is:

[0112]

[0113] In the formula, the total number of plant type categories is C, i represents the i-th peanut plant type, and TP i TN refers to the number of peanut plants that are actually of type i and are predicted to be of that type. i FP represents the number of peanut plants that do not actually belong to the i-th plant type but are predicted to belong to other plant types. i FN represents the number of peanut plants that do not actually belong to the i-th type but are misclassified as belonging to the i-th type. i This represents the number of peanut plants that actually belong to the i-th plant type but were misclassified as other plant types. (Accuracy) iThe prediction accuracy of the i-th type of peanut plant

[0114] Weighted precision is an important metric for measuring the predictive accuracy of a model across different plant types, effectively reflecting the model's overall predictive performance under uneven class distribution. Weighted precision considers the proportion of samples in each class while calculating the precision for each class, thus avoiding overall evaluation bias caused by fluctuations in smaller classes. This metric is calculated by... i Weighted precision (WPP) focuses on the model's actual performance across a large sample size and serves as an important reference for evaluating the model's overall discriminative ability. The formula for calculating WPP is as follows:

[0115]

[0116] In the formula ω i Let n be the proportion of class i in all samples. i This represents the number of samples in this category. The total number of strain types is C, and the number of true positives (TP) for the i-th strain type is [missing information]. i This refers to the number of peanut plots that are actually of type i and correctly predicted by the model to be of that type. False positives (FP) for type i plant type. i This represents the number of plots that do not actually belong to the i-th category but are misclassified as belonging to the i-th category by the model. This type of error is also known as Type I Error.

[0117] Weighted recall represents a model's overall ability to identify true samples across all plant types, reflecting the model's comprehensive level of avoiding false negatives in multi-class classification tasks. Compared to precision, which focuses more on the correctness of predictions, recall emphasizes the model's "detection ability" of positive samples. Weighted recall is calculated by weighting the recall of each class by sample size, taking into account the model's performance and class distribution across different classes, effectively mitigating evaluation bias caused by class imbalance. This metric is suitable for plant type identification tasks with high accuracy requirements, uneven sample distribution, or subtle structural differences between classes, and has significant reference value in model performance analysis and optimization. The formula for calculating Weighted Recall is:

[0118]

[0119] In the formula ω i Let n be the proportion of class i in all samples. i This represents the number of samples in this class. Recall i Let be the recall rate of the i-th strain type, and let C be the total number of strain types. Let be the true positive rate (TP) of the i-th strain type. iThis refers to the number of peanut plots that are actually of type i and that the model correctly predicts as belonging to that type; the false negative for type i plant type (FN) i The number of plots that actually belong to type i are misclassified by the model as other plant types. This error is also known as a Type II error.

[0120] Weighted F1 score 1-score F-squared is one of the important metrics for measuring the overall performance of a multi-class classification model, comprehensively considering the balance between precision and recall. Compared to evaluating precision or recall alone, F-squared... 1-score It focuses more on the trade-off between accuracy and completeness in positive class identification, and is particularly suitable for imbalanced class samples. In multi-class scenarios, the weighted F1 score will affect the F1 score of each class. 1-score Assigning weights to classes within the dataset reflects the model's overall discriminative ability across categories, balancing predictive ability and class importance. This is a crucial indicator for evaluating model stability and generalization ability. Weighted F1 score 1-score The calculation formula is as follows:

[0121]

[0122] In the formula ω i Let n be the proportion of class i in all samples. i This represents the number of samples in this category. The total number of plant type categories is C, Precision i Let Recall be the accuracy rate for the i-th plant type. i Let be the recall rate of the i-th strain type.

[0123] Step 4: Based on the growth fitting curve, time-series modeling of peanut canopy height and width is performed to obtain the dynamic change characteristics of canopy width-to-height ratio during the growth period. Key growth periods with plant type discrimination ability are identified from the time series, realizing accurate identification of peanut plant type based on canopy structure perception.

[0124] The Richards growth fitting curve model fully leverages its advantages in structural dynamics modeling to achieve continuous temporal sensing of key canopy structural parameters such as peanut canopy height and width. This model exhibits good flexibility and generalization ability, capable of fitting asymmetric changes in crop growth trends in actual farmland environments, demonstrating strong adaptability. By extracting daily structural parameters from the fitting results and combining them with statistical analysis methods, the key time windows most discriminative for plant type differentiation during the growth period can be further identified.

[0125] Step 401: To characterize the dynamic changes in peanut canopy structure parameters throughout the entire growth cycle, this study introduced the Richards growth model, performing nonlinear fitting on canopy height and width to obtain daily aspect ratio data. The Richards model has a flexible curve structure; by introducing shape adjustment parameters, it can effectively express asymmetric growth patterns during crop growth, such as... Figure 4 As shown. Among them Figure 4 In the figure, (1) represents the fitted curve of the growth of the upright canopy width. Figure 4 In the figure, (2) represents the fitted curve of the upright canopy height growth. Its calculation formula is:

[0126]

[0127] In the formula, y(t) represents the growth amount (canopy coverage) at time t. S is the maximum value of the model, that is, the maximum growth value that the system can achieve (the upper limit of canopy coverage). r is the growth rate, representing how fast the growth is. A is a constant related to the initial conditions, representing the relative growth amount in the initial state. v is a shape parameter, determining the shape of the curve, especially the steepness and symmetry of the curve. t is the time variable.

[0128] Step 402: In order to identify the key time window with the most discriminative power for plant type in the growth cycle, this study conducted a statistical test on the significant differences in the daily width-to-height ratio among different peanut plant types. The ANOVA analysis method was used to calculate the F value and p value on a daily basis to determine the specific date window in the growth cycle where peanut plant type can be predicted based on the width-to-height ratio.

[0129] To further illustrate the effects of the embodiments of the present invention, the following experiments were conducted: First, ANOVA analysis was used to determine the feasibility of peanut canopy perception-width-to-height ratio-driven plant type classification. From day 34 to day 70 after sowing, the F values ​​for all growth stages were significantly higher, while the p value was <0.05. This indicates that during this period, there were significant differences in the width-to-height ratio of peanut plants among different plant types, suggesting that the width-to-height ratio of peanut plants can be used as an effective canopy structure parameter to distinguish different growth types.

[0130] Peanut plant types were classified using the DT model, with the aspect ratio as the distinguishing feature. The classification performance remained relatively stable from day 20 to day 63 post-sowing, with an accuracy rate consistently above 95%, indicating that the aspect ratio has a strong discriminative ability in peanut plant type classification. Notably, the decision tree model achieved its best classification results for peanut plant types between day 42 and day 54 post-sowing, with an accuracy rate as high as 97%–98%. This demonstrates that during this growth period, the canopy structure perception-aspect ratio approach can effectively identify peanut plant types.

[0131] Finally, ANOVA analysis was used to evaluate the feasibility of WHR in predicting plant architecture at different growth stages. Figure 5 It can be seen that the predictive ability of WHR for plant architecture changes linearly over time. From day 20 to 24 post-sowing, the P-value is above 0.05, and the F-value is close to 0, indicating that WHR has low discriminative power for plant architecture and weak predictive ability. From day 25 to day 74 post-sowing, the P-value remains below 0.05, indicating that WHR has a strong explanatory power for plant architecture during this period, making it a suitable window for predicting peanut plant architecture based on WHR. However, after day 74 post-sowing, the P-value is above 0.05, and the F-value returns to near 0, indicating that WHR no longer has the ability to predict plant architecture. Notably, from day 46 to 52 post-sowing, the F-value reaches 250+, and the P-value is infinitely close to 0, indicating that plant architecture prediction based on peanut WHR is most reliable during this period.

[0132] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A low-altitude remote sensing monitoring method for peanut plant architecture driven by canopy structure perception, characterized in that, include: S1. Based on high-resolution RGB images, complete POS alignment and image reconstruction, generate digital surface model (DSM) and digital elevation model (DEM), and construct canopy height model (CHM) to estimate canopy height; S2. Quantitatively extract the visual features of peanut canopy and soil background, dynamically remove weed interference by DBSCAN clustering and estimate canopy width; S3. Based on the aspect ratio features extracted from canopy structure perception, construct a decision tree classification model based on aspect ratio to achieve automatic identification of different peanut plant types; One-way ANOVA was used to test the significance of peanut canopy width-to-height ratio in plant type identification, and the F-value was used to measure the relative strength of differences between groups and within groups. for: ； In the formula: ； ； in, For the mean square between groups, Within-group mean square For the first The number of samples in the group For the first group mean The population mean Number of categories; The total number of samples, Indicates the first The first in the group One observation value; When the means of different groups are exactly the same, the p-value of the current observation is obtained. for: ； In the formula, It has degrees of freedom. and Nk Distribution, with a significance level of 0.05 as the criterion, If the aspect ratio is less than 0.05, it proves that the peanut plant type can be determined by the aspect ratio. The DT algorithm is used to complete a multi-class classification task based on aspect ratio. The feature space is recursively partitioned to minimize the Gini coefficient. Using the objective function, classification nodes are generated layer by layer to achieve automatic identification of different peanut plant types; ； In the formula, For the first The proportion of class samples in the current node; S4. Based on the growth fitting curve, time-series modeling of peanut canopy height and width is performed to obtain the dynamic change characteristics of canopy width-to-height ratio during the growth period. Key growth periods with plant type discrimination ability are identified from the time series, realizing accurate identification of peanut plant type based on canopy structure perception.

2. The low-altitude remote sensing monitoring method for peanut plant architecture driven by canopy structure perception according to claim 1, characterized in that, In step S1, based on the first Digital surface model of a small peanut-growing area With digital elevation model Calculation of pixel-by-pixel difference Canopy height model of a small peanut area The calculation formula is as follows: ； in, This corresponds to the peanut-growing area plot.

3. The low-altitude remote sensing monitoring method for peanut plant architecture driven by canopy structure perception according to claim 1, characterized in that, In step S2, effective separation of the peanut canopy and soil is performed based on the color index EXG. First, the three color channels of the RGB image are normalized, as shown in the expression: ； in, ； ； ； In the formula, These are the channel values ​​for the red, green, and blue channels, respectively. The green band of the image, The red band of the image, The blue band of the image.

4. The low-altitude remote sensing monitoring method for peanut plant architecture driven by canopy structure perception according to claim 3, characterized in that, Based on the exponential image of the color index (EXG), a fixed threshold method was used to binarize the exponential image to obtain the peanut canopy region mask image at each growth stage. The expression is: ； In the formula, Indicates the first One reproductive period, Indicates pixel position, For pixels In the The growth stage was identified as the peanut canopy region. For pixels In the The growing season was identified as soil; The pixel position of the t-th reproductive period EXG color index at the location; Subsequently, canopy mask images were analyzed at each reproductive stage. Perform pixel-level logical AND operations to generate a cumulative temporal mask. : ； In the formula, For the cumulative phase mask, It is the sum of all reproductive periods. In the first A binary mask image generated during each reproductive period. For pixels In the Each reproductive period Identified as a vegetated area For pixels In the Areas not identified as vegetation during their growth period. The row and column where the pixel is located.

5. The low-altitude remote sensing monitoring method for peanut plant architecture driven by canopy structure perception according to claim 4, characterized in that, The DBSCAN clustering algorithm is used to cluster connected canopy pixels, dividing the extracted single-temporal peanut canopy mask into different pixel clusters; combined with the cumulative temporal mask... The algorithm determines whether each pixel cluster persists throughout the entire growth period of the peanut plant. If the current pixel cluster intersects with the cumulative phase mask, then the pixel cluster is a peanut canopy region; otherwise, the pixel cluster is determined to be a weed region, and the weeds are removed. This process is repeated to mark and remove weeds throughout the entire growth period of the peanut plant.

6. The low-altitude remote sensing monitoring method for peanut plant architecture driven by canopy structure perception according to claim 1, characterized in that, In step S4, the Richards growth fitting model is introduced to perform time-series fitting on key structural indicators of canopy height and width. ； In the formula, Indicates the first Growth during each reproductive period Let r be the maximum value of the model, r be the growth rate, and A be a constant. For shape parameters.

7. A low-altitude remote sensing monitoring system for peanut plant architecture driven by canopy structure perception, characterized in that, The method for implementing the peanut plant type low-altitude remote sensing monitoring method driven by canopy structure perception as described in any one of claims 1-6 includes: a peanut canopy height structure perception and processing module, a peanut canopy width structure perception and processing module, a peanut plant type identification module, and a peanut plant type accurate temporal sequence identification and window determination module. The peanut canopy height structure perception and processing module completes POS alignment and image reconstruction based on high-resolution RGB images, generates a digital surface model (DSM) and a digital elevation model (DEM), and constructs a canopy height model (CHM). The peanut canopy width structure perception and processing module quantitatively extracts the visual features of the peanut canopy and soil background, dynamically removes weed interference and estimates the canopy width through DBSCAN clustering; The peanut plant type identification module constructs a decision tree classification model based on the aspect ratio features extracted from canopy structure perception to achieve automatic identification of different peanut plant types. The peanut plant type accurate temporal identification and window determination module performs temporal modeling of peanut canopy height and canopy width based on growth fitting curves, obtains the dynamic change characteristics of canopy width-to-height ratio during the growth period, identifies key growth periods with plant type discrimination ability from the time series, and realizes accurate identification of peanut plant type based on canopy structure perception.

Images