Crop canopy coverage high-throughput acquisition method and system based on remote sensing of unmanned aerial vehicle

Abstract

The invention belongs to the technical field of low-altitude unmanned aerial vehicle remote sensing and precision agriculture, and discloses a crop canopy coverage high-throughput acquisition method and system based on unmanned aerial vehicle remote sensing. The method comprises the following steps: quantitatively describing a visual feature mode between a crop canopy and a soil background from a high-resolution RGB image; automatically mining association mapping between category semantics and visual feature modes by using a support vector machine method to obtain a preliminary extraction result of crop canopies and soil backgrounds, and generating a single-temporal crop canopy mask map; the method comprises the following steps of: optimizing all single-time-phase crop canopy mask patterns of crop growth by using a dynamic weed elimination method of an RGB (Red, Green, Blue) image time sequence, and respectively counting the number of peanut canopy pixel clusters and the number of soil pixel clusters in each plot cell in each plot to finish accurate estimation of the crop canopy coverage. The invention provides a basic technical means for the fields of agricultural precise management, crop high-throughput phenotype monitoring, digital breeding and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of low-altitude UAV remote sensing and precision agriculture, and discloses a method and system for high-throughput acquisition of crop canopy coverage by UAV remote sensing. Background Technique

[0002] The research of crop phenotype is usually realized by analyzing its parameter characteristics, such as canopy coverage, leaf area index, plant height, etc. A large number of studies have proved that there is a significant correlation between crop canopy coverage and other phenotypic trait parameters. Therefore, the accurate extraction research of crop canopy coverage is very important for crop phenotypic trait analysis and precise identification of germplasm resources. Crop canopy coverage (FVC) refers to the ratio of the vertical projection area of the crop canopy within the plot area to the total soil area, that is, the vegetation-soil ratio. In the field of agronomy, crop canopy coverage is an important parameter to describe the distribution of surface crops, which can reflect the light interception ability of crop leaves, judge the growth health status of crops and predict the biological yield.

[0003] At present, the estimation methods of crop canopy coverage mainly include two types: traditional ground measurement method and remote sensing monitoring method. The commonly used ground measurement method is the grid method. Although it has high accuracy, it has low efficiency and is difficult to meet the large-scale crop trait investigation. In recent years, UAV remote sensing technology has attracted more and more attention due to its advantages of high accuracy, low application cost and strong timeliness. And a large number of scholars have conducted many studies on using UAV remote sensing technology to extract crop canopy coverage. For the remote sensing monitoring method, vegetation-soil separation is the first and most important step for accurate extraction of crop canopy coverage. By classifying the image into two types of vegetation pixels and non-vegetation pixels, the proportion of vegetation pixels is directly calculated after classification. In images with low spatial resolution such as satellites, a single pixel may contain vegetation and non-vegetation information (mixed pixel), and it is impossible to accurately classify the pixel as vegetation or non-vegetation. Therefore, generally, the pixel dichotomy model is used to obtain the proportion of the vegetation component of the pixel by solving the proportion of each component of vegetation or non-vegetation in the mixed pixel.

[0004] Different from images with low spatial resolution such as those taken by satellites, the images obtained by low-altitude drones have a higher spatial resolution, which can reach 1 mm / pixel. Therefore, the canopy coverage can be estimated by directly counting vegetation pixels and non-vegetation pixels. In the field of agronomy, the segmentation of vegetation and soil in real farmland scenarios is often challenged by factors such as unstructured backgrounds, lighting changes, and weak contrast between crop backgrounds in complex field backgrounds, which has become a hot issue in current research. Therefore, agricultural researchers have conducted a large number of studies on crop canopy segmentation. Currently, the two main types of mainstream methods are threshold segmentation and machine learning methods. Although the threshold-based method is simple, intuitive, easy to understand and implement, this type of method requires dynamically determining the segmentation threshold. In the field of machine learning, there are often different visual patterns and presentation rules between crop canopies and backgrounds such as soil. Based on quantitatively describing the specific visual patterns between them, machine learning establishes a statistical model to automatically mine the association mapping between category semantics and visual patterns, and is widely used in land cover classification and target recognition in fields such as computer vision and precision agriculture. Therefore, the use of machine learning methods for vegetation and soil segmentation has received increasing attention.

[0005] Through the above analysis, the problems and defects of the existing technology are as follows: The existing technology uses methods such as threshold segmentation, machine learning, or deep learning to estimate crop coverage. However, these methods do not consider the impact of weeds on crop coverage estimation (especially in the seedling stage) in farmland scenarios, resulting in overestimated estimation results. And in the existing technology for removing the impact of weeds on crop canopy coverage estimation from high-resolution RGB images of drones, the accuracy of high-throughput acquisition of crop canopy coverage is low. Summary of the Invention

[0006] To overcome the problems existing in the related technology, the disclosed embodiments of the present invention provide a method and system for high-throughput acquisition of crop canopy coverage by drone remote sensing, specifically related to a technology for high-throughput acquisition of crop canopy coverage by low-altitude drone remote sensing.

[0007] The technical solution is as follows: A method for high-throughput acquisition of crop canopy coverage by drone remote sensing includes the following steps:

[0008] S1, quantitatively describe the visual feature patterns between the crop canopy and the soil background from high-resolution RGB images;

[0009] S2, use the support vector machine method to automatically mine the association mapping between category semantics and visual feature patterns, obtain the preliminary extraction results of the crop canopy and the soil background, generate a single-temporal crop canopy mask image, where the crop canopy area is marked as 1 and the soil background area is marked as 0;

[0010] S3. Using the dynamic weed removal method based on the RGB image time series, optimize the crop canopy mask maps of all single time phases of crop growth, determine the peanut canopy area and weed area, and mark and remove the weeds in the entire growth period of the crop.

[0011] S4. Based on the optimized results, separately count the number of peanut canopy pixel clusters and soil pixel clusters in each plot in each plot to accurately estimate the crop canopy coverage.

[0012] In step S1, the crop is peanut. Among the visual feature patterns that quantitatively describe the crop canopy and the soil background from the high-resolution RGB image, two color indices, the Excess Green Index (ExG) and the Color Index for Vegetation Extraction (CIVE), are used to separate the visual feature patterns of the peanut canopy and the soil background.

[0013] Furthermore, the Excess Green Index (ExG) is used to calculate the greenness degree of the crop, and the calculation formula is:

[0014] ExG = 2g - r - b (1)

[0015] In the formula, ExG is the Excess 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.

[0016] The calculation formulas for r, g, and b are:

[0017]

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

[0019] The Color Index for Vegetation Extraction (CIVE) is used to separate the green vegetation from the soil background, and the calculation formula is:

[0020] CIVE = 0.441R - 0.811G + 0.385B + 18.78745 (3)

[0021] In the formula, CIVE is the Color Index for Vegetation Extraction.

[0022] In step S2, use the support vector machine method to automatically mine the association mapping between the category semantics and the visual feature patterns, including:

[0023] Use the scikit-learn library in the Python environment to build an SVM classifier model, and use the SVC function in the scikit-learn library to train the training sample data. The call form of this function is:

[0024] svm_model = SVC('kernel') (4)

[0025] In the formula, svm_model is the defined support vector machine training model, SVC() indicates that the called one is the support vector machine classification in the support vector machine, and 'kernel' is the kernel function of this model;

[0026] Train a support vector machine (SVM) model, use the RGB values of pixels as feature inputs, and the classification labels as output targets; use the SVM model to replace the logic of the threshold method, input the RGB values of each pixel point into the SVM model, and output the classification labels; utilize the established SVM classifier model to classify the RGB images of peanuts at each growth stage obtained, and obtain the preliminary extraction results of the peanut canopy and the background such as the soil, and obtain a mask image that only contains the peanut images of the processed stage.

[0027] In step S3, the dynamic weed removal method using the RGB image time series includes:

[0028] (1) Cumulative temporal mask map generation, multiply the mask data of all growth stages pixel by pixel, and find the intersection of the crop canopy areas to obtain the cumulative temporal mask map;

[0029] (2) Combine the cumulative temporal mask map and perform DBSCAN clustering analysis to determine whether each pixel cluster persists throughout the entire growth period of the crop.

[0030] In step (1), the cumulative temporal mask map is obtained, and the expression is:

[0031]

[0032] In the formula, M is the cumulative temporal mask map Mask, T is the sum of time points, M t is the binary mask map generated at time point t, means that the pixel point (i, j) is recognized as a vegetation area at time point t, means that the pixel point (i, j) is not recognized as a vegetation area at time point t, and i, j are the row and column where the pixel is located.

[0033] In step (2), combine the cumulative temporal mask map and perform DBSCAN clustering analysis, including:

[0034] Use the DBSCAN clustering algorithm to cluster the connected canopy pixels, and divide the extracted single-temporal peanut canopy mask map into different pixel clusters; combine the cumulative temporal mask map to determine whether each pixel cluster persists throughout the entire growth period of the peanuts; if the current pixel cluster has an intersection with the cumulative temporal mask map, then this pixel cluster is the peanut canopy area; otherwise, it is determined that this pixel cluster is a weed area, and the weeds are removed; repeat this process to mark and remove the weeds throughout the entire growth period of the peanuts.

[0035] In step S4, accurate estimation of the crop canopy coverage is completed, and the expression is:

[0036]

[0037] In the formula, FVC (i,j) is the canopy coverage of the plot cell in the i-th row and j-th column, M (i,j) is the number of crop canopy pixel clusters in the plot cell in the i-th row and j-th column, N (i,j) is the number of soil pixel clusters in the plot cell in the i-th row and j-th column.

[0038] Another object of the present invention is to provide a high-throughput acquisition system for crop canopy coverage by unmanned aerial vehicle remote sensing. This system implements the high-throughput acquisition method for crop canopy coverage by unmanned aerial vehicle remote sensing. This system includes:

[0039] A visual feature pattern acquisition module, which is used to quantitatively describe the visual feature pattern between the crop canopy and the soil background from high-resolution RGB images;

[0040] A single-temporal crop canopy mask map generation module, which is used to automatically mine the association mapping between category semantics and visual feature patterns by using the method of support vector machine, obtain the preliminary extraction results of the crop canopy and the soil background, and generate a single-temporal crop canopy mask map. The crop canopy area is marked as 1, and the soil background area is marked as 0;

[0041] A dynamic weed removal module for the RGB image time series, which is used to optimize all single-temporal crop canopy mask maps during crop growth by using the dynamic weed removal method for the RGB image time series, determine the peanut canopy area and the weed area, and mark and remove the weeds throughout the growth period of the crop;

[0042] A crop canopy coverage estimation module, based on the optimized results, respectively counts the number of peanut canopy pixel clusters and the number of soil pixel clusters in each plot cell in each plot, and completes the accurate estimation of the crop canopy coverage.

[0043] Furthermore, the high-throughput acquisition system for crop canopy coverage by unmanned aerial vehicle remote sensing is carried on a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the functions in the above high-throughput acquisition system for crop canopy coverage by unmanned aerial vehicle remote sensing can be realized.

[0044] Combining all the above technical solutions, the beneficial effects of the present invention are as follows: The high-resolution RGB images obtained by the low-altitude remote sensing technology of the unmanned aerial vehicle in the present invention make it simple and easy to estimate the crop coverage by separating the crop canopy and non-vegetation backgrounds such as the soil.

[0045] The present invention utilizes the high-spatial-resolution image acquisition capabilities of low-altitude unmanned aerial vehicle (UAV) remote sensing, which are efficient and low-cost, and combines the advantages of computer vision in feature extraction. Through the time-series images of peanut canopy masks, information fusion in the time dimension is achieved, significantly enhancing the robustness of peanut field weed removal, realizing reliable separation of the planted soil, and further improving the accuracy and quality of crop canopy coverage estimation. The high-throughput acquisition technology for crop canopy coverage traits designed based on low-altitude UAV remote sensing in the present invention improves the automation and intelligent level of crop canopy coverage estimation, providing a basic technical means for fields such as agricultural precision management, crop high-throughput phenotyping monitoring, and digital breeding.

[0046] The present invention can achieve efficient acquisition of crop canopy coverage: By combining unmanned aerial vehicle (UAV) remote sensing with high-resolution RGB images, the present invention can rapidly acquire crop canopy coverage data in a high-throughput and automated manner, significantly improving the efficiency and accuracy compared with traditional manual measurement or low-resolution image analysis. The present invention can optimize agricultural production: Accurately monitor the growth of crops, provide data support for precision agriculture, optimize farmland management decisions (such as fertilization, irrigation, and weeding), and increase crop yields. The present invention can reduce the cost of weed removal: Through an automated dynamic weed removal method, the cost of manual weeding is significantly reduced, and the risk of crop yield reduction caused by weed competition is decreased. The present invention can achieve dynamic weed removal: A time-series-based dynamic weed removal method is proposed, which combines DBSCAN clustering and cumulative temporal mask analysis to achieve precise differentiation between crops and weeds. This innovative method is significantly superior to traditional single-temporal static analysis methods. The present invention can extract visual feature patterns combining multiple indicators: The present invention comprehensively uses the excess green index (ExG) and the color index for vegetation extraction (CIVE), and for the first time, based on high-resolution RGB images, models the characteristics of the crop canopy and soil background from multiple angles, improving the accuracy of identification. The present invention can achieve high-throughput precise monitoring: Traditional methods for monitoring canopy coverage rely mostly on manual surveys or medium- and low-resolution images. The present invention, through UAV high-resolution images and machine learning algorithms, realizes the efficient and automated acquisition of crop canopy coverage. The present invention can precisely distinguish between the crop canopy and weed areas: The high visual similarity between weeds and crops and their dynamic characteristics during the growth cycle make it difficult for traditional methods to accurately identify them. The present invention, by combining time-series analysis and density clustering algorithms, for the first time achieves dynamic weed removal. The present invention can overcome the low-efficiency bottleneck in rapid acquisition of coverage: Traditional canopy coverage monitoring is limited by the accuracy of low-resolution images and the efficiency of manual analysis, and it is difficult to be efficiently applied in large-scale farmland. The present invention significantly improves the processing speed and applicability. The present invention can achieve robustness in growth stage monitoring: Previous technologies rely strongly on single-temporal data and are easily affected by local errors. The present invention, through the dynamic method of generating cumulative temporal mask images, significantly enhances the stability and accuracy of data analysis.

[0047] Traditional methods tend to analyze crop canopy coverage through single-temporal images, without considering the dynamics of crop growth. The time-series dynamic analysis method proposed by the present invention effectively overcomes this limitation. Low utilization rate of high-resolution images: Previous studies mostly relied on low-resolution remote sensing images for large-area monitoring, believing that high-resolution images were difficult to apply efficiently due to large data volume and complex processing. The present invention gives full play to the potential of high-resolution images through algorithm optimization and automated processing procedures. Fixed thinking in weed identification: Traditional methods usually distinguish weeds by fixed thresholds or static feature extraction. The present invention combines clustering analysis and temporal mask maps to break through this fixed thinking and dynamically adjusts the analysis model to adapt to the characteristics of different growth stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure;

[0049] Figure 1 is a flowchart of a high-throughput method for obtaining crop canopy coverage by unmanned aerial vehicle remote sensing provided by an embodiment of the present invention;

[0050] Figure 2 is a schematic diagram of the principle for generating a cumulative temporal mask map provided by an embodiment of the present invention;

[0051] Figure 3 is a schematic diagram of a high-throughput system for obtaining crop canopy coverage by unmanned aerial vehicle remote sensing provided by an embodiment of the present invention, where (a) is an image map of an unprocessed plot, (b) is an image map of the plot with manually marked weed areas, (c) is an image map of the plot with weed areas marked using the present invention patent, (d) is a binary image map of the plot without weed removal, (e) is a binary image map of the plot with manually removed weeds, and (f) is a binary image map of the plot with weeds removed using the method of the present invention patent;

[0052] In the figure: 1. Visual feature pattern acquisition module; 2. Single-temporal crop canopy mask map generation module; 3. Dynamic weed removal module for RGB image time series; 4. Crop canopy coverage estimation module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention is provided with reference to the drawings. Many specific details are set forth in the following description to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0054] The innovation of the present invention lies in: by combining the visual feature patterns of high-resolution RGB images, the excess green index ExG, and the color index for vegetation extraction CIVE, the present invention automatically mines the mapping between class semantics and visual feature patterns using a support vector machine. A time series analysis method for dynamic weed removal is proposed, and DBSCAN clustering is combined to accurately distinguish the crop canopy and weed areas. Finally, the coverage estimation is optimized through the cumulative phase mask map, improving the efficiency, accuracy, and automation level of canopy coverage monitoring, and is applicable to precision agriculture and large-scale farmland management.

[0055] Example 1, as Figure 1 shown, the method for high-throughput acquisition of crop canopy coverage by unmanned aerial vehicle remote sensing provided in the embodiment of the present invention includes the following steps:

[0056] S1, quantitatively describe the visual feature pattern between the crop canopy and the soil background from high-resolution RGB images;

[0057] S2, automatically mine the association mapping between class semantics and visual feature patterns using a support vector machine method to obtain a preliminary extraction result of the crop canopy and the soil background, generate a single-phase crop canopy mask map, mark the crop canopy area as 1, and mark the soil background area as 0;

[0058] S3, use the dynamic weed removal method for the RGB image time series to optimize all single-phase crop canopy mask maps during crop growth, determine the peanut canopy area and weed area, and mark and remove the weeds throughout the growth period of the crop;

[0059] S4, based on the optimized result, respectively count the number of peanut canopy pixel clusters and soil pixel clusters in each plot in each plot to complete the accurate estimation of the crop canopy coverage.

[0060] Example 2, the present invention takes peanuts as the research object, uses the unmanned aerial vehicle low-altitude remote sensing technology to obtain a high-resolution RGB image sequence, extracts the canopy area of peanuts at different growth stages, and designs a method for high-throughput acquisition of crop canopy coverage by unmanned aerial vehicle remote sensing (a method for high-throughput acquisition of crop canopy coverage by low-altitude unmanned aerial vehicle remote sensing). Specifically include:

[0061] Step 1, data set source.

[0062] The present invention uses the UAV image data collected during different growth stages of peanuts. The collection of the UAV image data of peanuts was carried out in a peanut experimental field in a certain county, and the label files were manually marked in labelme. A certain county belongs to a typical warm temperate semi-humid continental climate, with distinct seasons, rain and heat in the same period, and the average annual temperature is about 14°C. The average altitude of a certain county is 85 meters, the average annual sunshine hours is 2759.1h, the average annual precipitation is 901.4mm, and the average annual frost-free period is 199 days, which is suitable for the growth of peanuts. The data collection started on May 31, 2023 and ended on August 9, 2023, covering the emergence stage, seedling stage, flowering and pegging stage, and podding stage of peanuts. The experiment used a DJI Mavic 3 UAV equipped with a visible light camera to collect data at noon on sunny and cloudless days. The DJIGS Pro platform was used for route planning and flight display, with the side overlap set to 80% and the flight altitude set to 20m. In the DJI Terra software, strip stitching and radiometric correction were carried out to generate digital orthophotos, and the peanut images of each period were divided into plots and batch extracted using ArcGIS 10.4 software.

[0063] Step 2, image preprocessing.

[0064] Affected by the complex background of the experimental soil, the image contains a lot of noise interference. To remove the image noise pollution and improve the stability of features, the present invention uses Gaussian low-pass filtering to enhance the grayscale images of the R, G, and B channels of the image respectively. Through Gaussian low-pass filtering, the image can be smoothed, the high-frequency noise in the soil background can be removed, and the influence of small pixel changes on the result can be reduced in the subsequent calculation of vegetation indices. Finally, the processed grayscale images of the three channels are merged to obtain the denoising result.

[0065] Step 3, visual feature pattern extraction.

[0066] Color feature is the most basic visual feature in image retrieval and also the most widely used basic image attribute in computer vision. This is because compared with other visual features, color feature has less dependence on the size, direction, and perspective of the image itself and has better robustness. Color indices are calculated linearly or non-linearly based on RGB images and have been widely used in semantic segmentation of crop images. The reasonable use of color indices can enhance the visual contrast between the foreground and background and reduce the influence of lighting conditions, improving the accuracy of crop segmentation. In the present invention, two color indices, the Excess Green index (ExG) and the Color Index for Vegetation Extraction (CIVE), are used to separate the visual feature patterns of peanut canopies and soil backgrounds.

[0067] Specifically, it includes:

[0068] Step 301, Excess Green index.

[0069] The Excess Green index (ExG) emphasizes the greenness of crops. Many previous studies have shown that the use of this index can clearly outline the crop contour. The calculation of this index requires prior normalization of the color values of three channels, and the specific calculation formula is as follows:

[0070] ExG = 2g - r - b (1)

[0071] In the formula, ExG is the Excess 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;

[0072] The calculation formulas for r, g, and b are:

[0073]

[0074] In the formula, R, G, and B are the channel values of the red, green, and blue channels respectively;

[0075] The ExG formula of the present invention is simple to calculate, only relying on the linear combination of the green (G), red (R), and blue (B) channel values, with high calculation efficiency. Moreover, it has a good effect on distinguishing vegetation areas with significant color characteristics, especially suitable for scenes with obvious greenness. By emphasizing the green channel and reducing the weights of red and blue, the vegetation is significantly higher than the background in terms of the index value. It can be directly applied to various optical image data (RGB images) without additional bands or complex preprocessing. For images with low resolution and obvious color distribution, ExG can better extract vegetation, which greatly reduces the requirements for equipment.

[0076] Step 302, extract the color index of vegetation.

[0077] The Color Index of Vegetation Extraction (CIVE) was proposed based on the research of soybean and beet fields, and it was found that CIVE is very effective in separating green vegetation from the outdoor soil background. The specific calculation formula is as follows:

[0078] The Color Index of Vegetation Extraction CIVE, used to separate green vegetation from the soil background, the calculation formula is:

[0079] CIVE = 0.441R - 0.811G + 0.385B + 18.78745 (3)

[0080] In the formula, CIVE is the Color Index of Vegetation Extraction.

[0081] In the present invention, compared with ExG, CIVE assigns different weights to the R, G, and B channels, more precisely differentiating vegetation areas and non-vegetation areas. It has a better vegetation extraction effect under complex backgrounds, can improve accuracy in various scenarios, and has strong background suppression ability. Under conditions of high light variation and scene complexity, CIVE is relatively more stable and less affected by light conditions.

[0082] Step 4, the method of the soil-plant separation model.

[0083] The method of machine learning can automatically learn effective features in data, improve the accuracy and robustness of classification, and a trained machine learning model usually has strong generalization ability and can achieve good classification results on unseen data, making it more reliable in practical applications. Therefore, the present invention uses the method of the classic machine learning model - support vector machine to establish the association mapping between category semantics and visual feature patterns. The present invention uses the scikit-learn (sklearn) library in the Python environment to construct an SVM classifier model (soil-plant separation model), and uses the SVC function in the library to train the training sample data. The calling form of this function is:

[0084] svm_model = SVC('kernel') (4)

[0085] In the formula, svm_model is the defined support vector machine training model, SVC() indicates that the called one is the support vector machine classification in the support vector machine, and 'kernel' is the kernel function of this model;

[0086] First, train a support vector machine (SVM) model, using the RGB values of pixels as feature inputs and classification labels (such as canopy and non-canopy) as output targets. Use the SVM model to replace the logic of the threshold method, input the RGB values of each pixel point into the SVM model, and output the classification label. Finally, use the established soil-plant separation model to classify the RGB images of peanuts at each growth stage obtained, obtain the preliminary extraction results of peanut canopy and soil and other backgrounds, and generate a single-temporal peanut canopy mask image, where the peanut canopy area is marked as 1 and the soil background area is marked as 0.

[0087] Step 5, removing weeds using time-series images.

[0088] Peanut fields are prone to weed growth before they close their rows (especially during the seedling stage). Since the visual feature patterns of weeds and peanut canopies are very similar, conventional soil-plant separation models are difficult to effectively distinguish between weeds and peanut canopies, especially in complex farmland scenarios, where this problem is even more prominent. To overcome this challenge, based on the characteristic that weeds do not exist in all growth stages of crops, the present invention designs a dynamic weed removal method using RGB image time series. Specifically, it includes:

[0089] Step 501, as Figure 2 shown, cumulative temporal mask map generation. Multiply the mask data of all growth stages pixel by pixel, that is, find the intersection of peanut canopy areas, to obtain a cumulative temporal mask map, and obtain the cumulative temporal mask map according to formula (5).

[0090]

[0091] In the formula, M is the cumulative temporal mask map Mask, T is the sum of time points, M t is the binary mask map generated at time point t, means that the pixel point (i, j) is recognized as a vegetation area at time point t, means that the pixel point (i, j) is not recognized as a vegetation area at time point t, and i, j are the row and column where the pixel is located.

[0092] This mask map can retain the areas that are shown as peanut plants in all growth stages, thus serving as a reference benchmark for the peanut plant area to effectively remove interfering objects (such as weeds and other noises) that appear in certain phases but do not persist.

[0093] DBSCAN clustering analysis.

[0094] For the single-phase peanut canopy mask map, the present invention uses the DBSCAN clustering algorithm to cluster the connected canopy pixels. DBSCAN is a density-based clustering algorithm that classifies data points into core points, border points, and noise points. Core points represent points with enough (usually at least) points within a certain radius. Border points represent points where the number of points within the radius is less than the requirement for core points but belong to at least one cluster. Noise points represent points that are neither core points nor border points. It identifies the attributes of points through the distance between points, and then divides the extracted single-phase peanut canopy mask map into different pixel clusters. Then, combined with the cumulative temporal mask map, it is judged whether each pixel cluster persists throughout the entire growth period of peanuts. If the current pixel cluster intersects with the cumulative temporal mask map, then this pixel cluster is the peanut canopy area; otherwise, it is determined that this pixel cluster is the weed area, thus achieving the removal of weeds. Repeat this process to mark and remove the weeds in the entire growth period of peanuts.

[0095] Step 6, estimating canopy coverage.

[0096] Based on the optimized results of the soil-plant separation model and the dynamic weed removal method using the RGB image time series, count the number of peanut canopy pixel clusters and the number of soil pixel clusters in each plot cell respectively, and calculate the canopy coverage of peanuts according to Equation (6).

[0097]

[0098] In the formula, FVC (i,j) is the canopy coverage of the plot cell in the i-th row and j-th column, M (i,j) is the number of crop canopy pixel clusters in the plot cell in the i-th row and j-th column, and N (i,j) is the number of soil pixel clusters in the plot cell in the i-th row and j-th column.

[0099] Example 3, as Figure 3 shown, the embodiment of the present invention provides a high-throughput acquisition system for crop canopy coverage by unmanned aerial vehicle remote sensing, including:

[0100] A visual feature pattern acquisition module 1 for quantitatively describing the visual feature pattern between the crop canopy and the soil background from high-resolution RGB images;

[0101] A single-temporal crop canopy mask map generation module 2 for automatically mining the association mapping between the category semantics and the visual feature pattern by using the support vector machine method, obtaining the preliminary extraction results of the crop canopy and the soil background, and generating a single-temporal crop canopy mask map, with the crop canopy area marked as 1 and the soil background area marked as 0;

[0102] A dynamic weed removal module 3 for the RGB image time series, which uses the dynamic weed removal method for the RGB image time series to optimize all single-temporal crop canopy mask maps of crop growth, determine the peanut canopy area and the weed area, and mark and remove the weeds in the entire growth period of the crop;

[0103] A crop canopy coverage estimation module 4, based on the optimized results, respectively count the number of peanut canopy pixel clusters and the number of soil pixel clusters in each plot cell in each plot, and complete the accurate estimation of the crop canopy coverage.

[0104] To further illustrate the relevant effects of the embodiments of the present invention, the following experiments are carried out: As Figure 3As shown in Figures (a)-(f) therein, the vegetation coverage of the binarized plot image without weed removal was calculated, and the plot vegetation coverage was 40.8227%. The calculation result of the vegetation coverage rate of the binarized plot after manually annotating the weed area was 36.5236%. For the binarized plot image obtained by using the method of this invention patent, its vegetation coverage was 36.5555%. This proves that this invention patent can accurately estimate the vegetation coverage well and has a positive effect in weed removal and peanut area recognition.

[0105] The above is only a relatively preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A method for obtaining crop canopy coverage by remote sensing using an unmanned aerial vehicle (UAV), characterized in that: The method comprises the following steps: S1, quantitative description of visual feature patterns between crop canopy and soil background from high-resolution RGB images; S2, using the support vector machine method to automatically mine the association mapping between category semantics and visual feature patterns, obtain the preliminary extraction results of crop canopy and soil background, and generate a single-phase crop canopy mask map, with the crop canopy area marked as 1 and the soil background area marked as 0; S3, using the dynamic weed removal method of RGB image time series, optimizes the crop canopy mask map of all single-phase crop growth, determines the peanut canopy area and weed area, and marks and removes weeds throughout the entire growth period of the crop; S4, based on the optimized results, count the number of peanut canopy pixel clusters and soil pixel clusters in each plot separately to complete the accurate estimation of crop canopy coverage.

2. The method for obtaining crop canopy coverage by UAV remote sensing according to claim 1, characterized in that: In step S1, the crop is peanut, and the visual feature pattern between the crop canopy and the soil background is quantitatively described from the high-resolution RGB image. Two color indices, namely the super green index ExG and the vegetation extraction color index CIVE, are used to separate the visual feature pattern between the peanut canopy and the soil background.

3. The method for obtaining crop canopy coverage by UAV remote sensing according to claim 2, characterized in that: The super green index ExG is used to calculate the greenness of crops. The calculation formula is: ExG=2g-rb (1) Where 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; The calculation formula of r,g,b is: In the formula, R, G, and B are the channel values ​​of the red, green, and blue channels respectively; The vegetation extraction color index is CIVE, which is used to separate green vegetation from the soil background. The calculation formula is: CIVE=0.441R-0.811G+0.385B+18.78745 (3) Where CIVE is the vegetation extraction color index.

4. The method for obtaining crop canopy coverage by remote sensing using an unmanned aerial vehicle according to claim 1, characterized in that: In step S2, the association mapping between category semantics and visual feature patterns is automatically mined using a support vector machine method, including: Use the scikit-learn library in the Python environment to build an SVM classifier model, and use the SVC function in the scikit-learn library to train the training sample data. The function is called in the form of: svm_model=SVC(′kernel′) (4) In the formula, svm_model is the defined support vector machine training model, FVC() indicates that the support vector machine classification in the support vector machine is called, and 'kernel' is the kernel function of the model; A support vector machine (SVM) model was trained, using the RGB values ​​of pixels as feature inputs and classification labels as output targets. The SVM model was used to replace the logic of the threshold method, and the RGB value of each pixel was input into the SVM model to output classification labels. The established SVM classifier model was used to classify the acquired RGB images of peanuts at each growth period, and preliminary extraction results of the background such as the peanut canopy and soil were obtained, thus obtaining a mask image that only contained the peanut images of the processed period.

5. The method for high-throughput acquisition of crop canopy coverage by UAV remote sensing according to claim 1, characterized in that: In step S3, a dynamic weed removal method using a time series of RGB images includes: (1) Generate a cumulative temporal mask map: multiply the mask data of all growth periods pixel by pixel, intersect the crop canopy area, and obtain a cumulative temporal mask map; (2) Combined with the cumulative temporal mask map, DBSCAN cluster analysis is performed to determine whether each pixel cluster persists throughout the entire growth period of the crop.

6. The method for obtaining crop canopy coverage by remote sensing using an unmanned aerial vehicle according to claim 5, characterized in that: In step (1), the cumulative phase mask is obtained, which is expressed as: Where M is the cumulative phase mask, T is the sum of the time points, and M t is the binary mask image generated at time point t, The pixel (i, j) is identified as a vegetation area at time point t. The pixel point (i, j) is not identified as a vegetation area at time point t, and i, j are the row and column where the pixel is located.

7. The method for obtaining crop canopy coverage by remote sensing using an unmanned aerial vehicle according to claim 5, characterized in that: In step (2), DBSCAN clustering analysis is performed in combination with the cumulative phase mask image, including: The DBSCAN clustering algorithm is used to cluster the connected canopy pixels, and the extracted single-phase peanut canopy mask map is divided into different pixel clusters; combined with the cumulative phase mask map, it is determined whether each pixel cluster persists in the entire growth period of peanut growth; if the current pixel cluster and the cumulative phase mask map have an intersection, the pixel cluster is the peanut canopy area; otherwise, the pixel cluster is determined to be a weed area, and the weeds are removed; the process is repeated to mark and remove the weeds throughout the growth period of peanut growth.

8. The method for high-throughput acquisition of crop canopy coverage by UAV remote sensing according to claim 1, characterized in that: In step S4, the crop canopy coverage is accurately estimated, and the expression is: In the formula, FVC (i,j) is the canopy coverage of the plot in the i-th row and j-th column, M (i,j) is the number of crop canopy pixel clusters in the plot with the i-th row and j-th column, N (i,j) is the number of soil pixel clusters in the plot with the i-th row and j-th column.

9. A high-throughput acquisition system for crop canopy coverage using UAV remote sensing, characterized in that: The system implements the high-throughput acquisition method of crop canopy coverage by drone remote sensing as described in any one of claims 1 to 8, and the system comprises: A visual feature pattern acquisition module (1) is used to quantitatively describe the visual feature pattern between the crop canopy and the soil background from the high-resolution RGB image; A single-phase crop canopy mask image generation module (2) is used to automatically mine the association mapping between category semantics and visual feature patterns using a support vector machine method, obtain preliminary extraction results of the crop canopy and soil background, and generate a single-phase crop canopy mask image, with the crop canopy area marked as 1 and the soil background area marked as 0; The dynamic weed removal module (3) of the RGB image time series is used to optimize the crop canopy mask map of all single-phase crop growth using the dynamic weed removal method of the RGB image time series, determine the peanut canopy area and the weed area, and mark and remove the weeds in the entire growth period of the crop growth; The crop canopy coverage estimation module (4) counts the number of peanut canopy pixel clusters and the number of soil pixel clusters in each plot in each plot based on the optimized results, and completes the accurate estimation of crop canopy coverage.

10. The crop canopy coverage high-throughput acquisition system for UAV remote sensing according to claim 9, characterized in that: The high-throughput acquisition system for crop canopy coverage using UAV remote sensing is carried on a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the functions of the high-throughput acquisition system for crop canopy coverage using UAV remote sensing can be realized.

Images