A method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles

Abstract

The present invention provides a method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles, which belongs to the field of cabbage cultivation technology, including: collecting a plurality of multispectral images of different varieties of Chinese cabbage in different growth stages based on unmanned aerial vehicles; generating digital orthophotos, digital surface models and dense three-dimensional point clouds of different varieties of Chinese cabbage by image stitching; using improved multispectral UNet combined with watershed algorithm and PointNet++ combined with density clustering algorithm to segment individual Chinese cabbage plants, and quantitatively obtain the spectral reflectance and three-dimensional point cloud phenotypic information of individual Chinese cabbage plants; combining a dual-input convolutional neural network model to construct a prediction model for relative chlorophyll content and nitrogen content, and establishing a leaf color discrimination model to monitor the growth of Chinese cabbage during the growth period. The present invention combines unmanned aerial vehicle multispectral imaging technology and deep learning to achieve quantitative acquisition of Chinese cabbage phenotypes and growth monitoring throughout the growth period, with a short sampling time, replacing time-consuming and labor-intensive methods such as manual visual inspection and ruler measurement.

Description

Technical Field

[0001] The present invention relates to the technical field of Chinese cabbage cultivation, and in particular to a method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles. Background Art

[0002] Chinese cabbage is an important leafy vegetable of the Brassica family. It is one of the most popular and frequently consumed vegetables in my country, and one of the most important vegetables in Asia. The height, plant shape, expansion, and leaf color of Chinese cabbage are important traits in Chinese cabbage breeding, which are related to the degree of cohesion and yield during the cabbage heading period and commercial quality. However, there are currently the following problems in the phenotypic determination and high-throughput acquisition of Chinese cabbage:

[0003] a. Traditional Chinese cabbage phenotyping is based on human judgment, which is highly subjective and cannot be accurately quantified;

[0004] b. The phenotyping methods of Chinese cabbage based on image processing are mostly laboratory-based, time-consuming and labor-intensive;

[0005] c. The Chinese cabbage phenotyping method based on drone imaging technology is mainly based on two-dimensional images, which cannot integrate multi-source data and the measured phenotypic traits are relatively single;

[0006] d. Currently, most studies on high-throughput acquisition of Chinese cabbage phenotypes are group studies, and it is impossible to perform high-throughput acquisition of multiple phenotypic traits such as height and expansion degree of individual cabbages at multiple growth stages. Summary of the invention

[0007] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a high-throughput acquisition and monitoring method of Chinese cabbage phenotypes based on multi-source data of unmanned aerial vehicles. Based on the combination of unmanned aerial vehicle multispectral imaging technology and deep learning, the quantitative acquisition of Chinese cabbage phenotypes and the growth monitoring of Chinese cabbage throughout the growth period are realized, and the sampling time is short. At the same time, the phenotypic traits of Chinese cabbage throughout the growth period can be accurately predicted, replacing time-consuming and labor-intensive methods such as manual visual inspection and measurement. It has the characteristics of fastness and accuracy, and provides a valuable reference for quickly acquiring phenotypic data of Chinese cabbage with different genotypes.

[0008] To achieve the above object, the present invention provides the following scheme: a method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles, comprising the following steps:

[0009] S1, using drones to collect images of multiple plots of different varieties of Chinese cabbage at different growth stages;

[0010] S2. Generate digital orthophotos, digital surface models and dense three-dimensional point clouds of different varieties of Chinese cabbage by image stitching based on the collected images of different growth stages;

[0011] S3, combining the acquired digital orthophoto, digital surface model and dense three-dimensional point cloud with deep learning to segment individual Chinese cabbage plants, and obtain two-dimensional and three-dimensional phenotypic information of individual Chinese cabbage plants;

[0012] S4. Construct a prediction model for chlorophyll and nitrogen content, a leaf color quantitative acquisition model, and a Chinese cabbage expansion degree quantitative acquisition model based on a dual-input convolutional neural network model to monitor the growth of Chinese cabbage during the growth period.

[0013] Preferably, in step S1, collecting a plurality of images of different varieties of Chinese cabbage at different growth stages based on a drone includes:

[0014] A drone equipped with a five-channel multispectral camera was used to obtain images of different varieties of Chinese cabbage at different growth stages, and multi-angle circular cross-route shooting was used to obtain image data from different angles.

[0015] Preferably, in step S2, the image splicing is generated in the following manner:

[0016] The collected images at different growth stages are integrated through UAV image synthesis to obtain a digital orthophoto of the entire field, and a dense three-dimensional point cloud and digital surface model of the plot are generated through a registration algorithm.

[0017] Preferably, in step S3, segmenting a single Chinese cabbage plant includes:

[0018] The images provided by the digital orthophotos were segmented into 640×640 sub-images through the repetition rate segmentation program, and Labelme was used to mark individual cabbages for the subsequent training of the UNet model. Cloucompare was used to segment the collected images into multiple sub-regions, and the point cloud of individual cabbages was marked for the subsequent training of the PointNet++ model. The phenotypic data of individual cabbages were obtained through two-dimensional and three-dimensional information.

[0019] Preferably, the two-dimensional and three-dimensional phenotypic information of a single Chinese cabbage plant includes but is not limited to plant height, plant width, spread and leaf color.

[0020] Preferably, in step S4, a prediction model for chlorophyll and nitrogen content based on a dual-input convolutional neural network model is constructed, comprising:

[0021] S41. Use the YOLOv8 transfer learning model to identify and locate Chinese cabbage in the seedling stage, and use QGIS software to combine target detection results and RTK technology to calibrate the position;

[0022] S42. Use the improved UNet semantic segmentation model combined with the watershed algorithm to segment a single cabbage plant to obtain a two-dimensional image segmentation result, and use the improved PointNet++ point cloud semantic segmentation model to perform complex background removal operations on the sub-regions of the three-dimensional point cloud, and combine the density clustering algorithm to achieve single cabbage plant segmentation, and obtain a single plant three-dimensional point cloud segmentation result;

[0023] S43, calculating the length and width pixel points of a single Chinese cabbage plant, a vegetation index image, and a multispectral image according to the two-dimensional image segmentation result;

[0024] S44, calculating the length, width and height of the bounding box of a single Chinese cabbage plant according to the segmentation result of the three-dimensional point cloud, and extracting an index for quantifying the spread of the Chinese cabbage canopy;

[0025] S45. Based on the dual-input convolutional neural network model, a prediction model for chlorophyll and nitrogen content was established.

[0026] Preferably, in step S4, a leaf color quantitative acquisition model is constructed, including:

[0027] Based on the multispectral image of a single Chinese cabbage plant, a leaf color quantitative acquisition model was established, and the average reflectance of different bands of a single Chinese cabbage plant was calculated and the spectral values ​​were combined to obtain different characteristic variables.

[0028] According to different characteristic variables and collected images of single Chinese cabbage, a dual-input convolutional neural network model was established for Chinese cabbage growth monitoring.

[0029] Preferably, in step S4, a Chinese cabbage expansion degree quantitative acquisition model is constructed, comprising:

[0030] The 3D point cloud parameters of a single Chinese cabbage plant are obtained by performing spherical fitting, circular fitting and projection processing on the 3D point cloud of a single Chinese cabbage plant, and different quantitative indicators are obtained by performing ratio combination on the 3D point cloud parameters of a single Chinese cabbage plant.

[0031] The principal component analysis was performed on the different quantitative indicators obtained, and the maximum and minimum normalization was performed on each principal component in combination with the contribution weight of each principal component to obtain the value of the comprehensive quantitative Chinese cabbage canopy expansion.

[0032] The extracted quantitative indicators of Chinese cabbage canopy expansion were subjected to multivariate stepwise regression analysis with the values ​​of comprehensive quantitative Chinese cabbage canopy expansion, a leaf color quantification model was established, and indicators with strong correlation were extracted for the quantification of plant expansion in the later stage.

[0033] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0034] The present invention provides a method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes based on multi-source data of unmanned aerial vehicles. Based on the combination of unmanned aerial vehicle multispectral imaging and deep learning, the method realizes quantitative acquisition of Chinese cabbage phenotypes and growth monitoring of Chinese cabbage throughout the growth period, with a short sampling time. At the same time, the phenotypic traits of Chinese cabbage throughout the growth period can be accurately predicted, replacing time-consuming and labor-intensive methods such as manual visual inspection and measurement. The method is fast and accurate, and provides a valuable reference for quickly acquiring phenotypic data of Chinese cabbage with different genotypes. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 A flowchart is provided for a method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles of the present invention;

[0037] Figure 2 A quantitative result diagram of the leaf color of Chinese cabbage at the rosette stage provided by an embodiment of the present invention;

[0038] Figure 3 The center point positioning and plant number statistics diagram of Chinese cabbage in the seedling stage provided by the embodiment of the present invention;

[0039] Figure 4 The Chinese cabbage plant spread and plant height prediction fitting diagram based on three-dimensional point cloud provided by the embodiment of the present invention, wherein: Figure 4 a is the predicted fitting diagram of Chinese cabbage plant length, Figure 4 b is the prediction fitting diagram of Chinese cabbage plant width, Figure 4 c is the predicted fitting diagram of Chinese cabbage plant height based on three-dimensional point cloud numerical calculation. Figure 4 d is the predicted fitting diagram of plant height calculated by DSM elevation of Chinese cabbage;

[0040] Figure 5 A schematic diagram of the change in plant length of Chinese cabbage in each period provided by an embodiment of the present invention;

[0041] Figure 6 A schematic diagram of the change of Chinese cabbage plant width in each period provided by an embodiment of the present invention;

[0042] Figure 7 A schematic diagram of the change of nitrogen content in Chinese cabbage in each period provided by an embodiment of the present invention;

[0043] Figure 8A schematic diagram of the change in relative chlorophyll content of Chinese cabbage plants in each period provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The purpose of the present invention is to provide a method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes based on multi-source data of unmanned aerial vehicles. Based on the combination of unmanned aerial vehicle multispectral imaging and deep learning, the method can realize the quantitative acquisition of Chinese cabbage phenotypes and the growth monitoring of Chinese cabbage throughout the growth period, with a short sampling time. At the same time, the phenotypic traits of Chinese cabbage throughout the growth period can be accurately predicted, replacing time-consuming and labor-intensive methods such as manual visual inspection and measurement. The method is fast and accurate, and provides a valuable reference for quickly acquiring phenotypic data of Chinese cabbage with different genotypes.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Embodiment 1

[0048] like Figure 1 As shown, the present invention provides a method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from drones, comprising the following steps:

[0049] S1. Using drones to collect images of multiple cabbages of different varieties at different growth stages. The collection dates and weather conditions are shown in Table 1. The real plant width, plant height, nitrogen content, and relative chlorophyll content are also collected. The data distribution at different stages is shown in Table 1. Figure 5-8 As shown, N1-N12 represent twelve varieties of cabbage;

[0050] Reference Figure 5 and Figure 6 , we can directly find the changes in the length and width of Chinese cabbage in each period, referring to Figure 7 , the change of nitrogen content has certain fluctuations, which is directly related to the fertilization of the plants; refer to Figure 8 , the relative chlorophyll SPAD of Chinese cabbage plants in each period was relatively close, and the change range was similar to the nitrogen content, proving that the change of SPAD was directly correlated with the nitrogen content.

[0051] S2. Generate digital orthophotos, digital surface models and dense three-dimensional point clouds of different varieties of Chinese cabbage by image stitching based on the collected images of different growth stages;

[0052] S3, combining the acquired digital orthophoto, digital surface model and dense three-dimensional point cloud with deep learning to segment individual Chinese cabbage plants, and obtain two-dimensional and three-dimensional phenotypic information of individual Chinese cabbage plants;

[0053] S4. Construct a prediction model for chlorophyll and nitrogen content, a leaf color quantitative acquisition model, and a Chinese cabbage expansion degree quantitative acquisition model based on a dual-input convolutional neural network model to monitor the growth of Chinese cabbage during the growth period.

[0054] Specifically, the obtained digital orthophotos at the seedling stage are used to detect and locate individual cabbages using a deep learning model, and the center point of each cabbage at the seedling stage is obtained to achieve subsequent position tracking;

[0055] Specifically, ENVI was used to obtain the spectral information of the outer leaves of Chinese cabbage with different color characteristics (yellow-green, light green, green, gray-green, and dark green) during the rosette stage, and the spectral information was combined to perform feature combination, and the Chinese cabbage color definition model was established by combining principal component analysis and multivariate linear regression.

[0056] Specifically, the digital orthophotos, digital surface models and dense three-dimensional point clouds obtained at the rosette stage and the heading stage are respectively segmented into individual Chinese cabbage plants using a deep learning model to obtain the phenotypic information of individual Chinese cabbage plants;

[0057] Wherein, in step S1, collecting images of the growth period of Chinese cabbage by using a drone includes:

[0058] A drone equipped with a five-channel multispectral camera was used to obtain images of different varieties of Chinese cabbage at different growth stages (one visible light image and five multispectral images). Multi-angle (90°, 180°, 270°, 360°) cross-route shooting was used to obtain image data at different angles to facilitate the subsequent generation of dense point clouds.

[0059] Secondly, in step S2, the image stitching is generated in the following way:

[0060] Through the UAV image synthesis method, the images collected at different growth stages are spliced ​​according to the POS points of each image to obtain the digital orthophoto of the entire field, and the dense three-dimensional point cloud and digital surface model of the plot are generated through the registration algorithm.

[0061] Again, the definition model for Chinese cabbage leaf color includes:

[0062] The spectral information of the outer leaves of Chinese cabbage with different color characteristics in the rosette period was extracted by ENVI software, and a total of 26 features were obtained by combining image features, such as R / (R+G+B), B / (R+G+B), RE / R, GB, R / B, H, S, V, etc. The principal component analysis was used again to extract the three principal components PC1, PC2, and PC3 with characteristic values ​​greater than 1, and the cumulative contribution rate was 98.65%. The sum of each principal component after standardization and multiplication by the weight of each principal component was obtained again as the value of quantified leaf color. This quantified value was normalized to the maximum and minimum and multiplied by 100 to achieve leaf color 0-100 quantization, and this quantified value was established with the original 26 features. A multivariate stepwise regression model was established to obtain the most color-related features NIR / R, B / (R+G+B) and G / B. Among them, the color quantification formula is:

[0063] Y=391.815-0.657*(NIR / R)-409.78*(B / (R+G+B))-17.376*(G / B);

[0064] This formula is used to quantify the color of each cabbage after segmentation, such as Figure 2 To quantify the results, the corresponding value below each quantification result represents the depth of leaf color. The larger the value, the greener the leaves, so that the color quantification value of each plant can be quickly obtained.

[0065] Again, the target detection of single cabbage seedlings includes:

[0066] The seedling stage Chinese cabbage was identified and located through the YOLOv8 transfer learning model, and the position was calibrated based on the QGIS software combined with the target detection results. The RTK recorded GPS position of each plant was found using each center point, so that the calibrated mask file runs through the entire growth period of the Chinese cabbage. The seedling stage positioning detection results are as follows: Figure 3 , the figure shows the center point of each cabbage. Figure 3 Substituting the results shown in into QGIS, the phenotypic tracking and analysis of Chinese cabbage plants over the entire period can be achieved;

[0067] Next, in step S3, segmenting the two-dimensional multispectral image of a single Chinese cabbage plant includes:

[0068] The digital orthophoto image was cut into 640×640 sub-images through the repetition rate segmentation program, and the image was labeled using LabelMe. The generated dense 3D point cloud was segmented into multiple sub-regions using CloudCompare, and each cabbage was labeled. The dataset obtained from the above steps was used to build a deep learning model in order to obtain the phenotypic information of a single cabbage plant.

[0069] Again, the segmentation of the 3D point cloud of a single Chinese cabbage plant includes:

[0070] A total of 33 indicators that can quantify the three-dimensional expansion degree are obtained through sphere fitting, projection and circle fitting, such as sphere radius, circle radius, length / width, etc., and principal component analysis is used again to extract seven principal components PC1, PC2, PC3, PC4, PC5, PC6, and PC7 with eigenvalues ​​greater than 1, with a cumulative contribution rate of 89.17%. The sum of each principal component after standardization and multiplication by the weight of each principal component is obtained again as the quantified value of the three-dimensional expansion degree. This quantified value is normalized to the maximum and minimum and then multiplied by 100 to achieve the quantification of the expansion degree 0-100. A multivariate stepwise regression model is established with this quantified value and the original 33 features, and 10 parameters most related to the expansion degree are obtained, including the circumference of the XOY projection circumscribed circle, height / length+width+height, sphere radius, sphere diameter, and circle diameter.

[0071] The two-dimensional and three-dimensional phenotypic information of a single Chinese cabbage plant includes but is not limited to plant height, plant width, spread and leaf color.

[0072] Finally, in step S4, a prediction model for chlorophyll and nitrogen content and a leaf color quantitative acquisition model based on a dual-input convolutional neural network model are constructed to monitor the growth of Chinese cabbage in the growth period, including:

[0073] S41. Use the improved UNet semantic segmentation model combined with the watershed algorithm to segment individual Chinese cabbage plants, including: segmenting each Chinese cabbage plant in the RGB image and multispectral image to obtain the two-dimensional image segmentation result. The segmentation model based on the multispectral image is the best, with a training set accuracy of 96.25% and a loss of 0.05, a validation set accuracy of 93.45% and a loss of 0.13, and the learning rate of the model is set to 1e -5 , the number of iterations epoch is 100. The improved PointNet++ point cloud semantic segmentation model combined with the density clustering algorithm is used to segment the sub-regions of the three-dimensional point cloud into individual cabbages, and the three-dimensional point cloud segmentation results are obtained. When 58 epochs are performed, the segmentation accuracy of the model is the highest, and the learning rate of the model is 1e -5 , the accuracy of the training set is 83.59%, the loss is 0.048, the accuracy of the validation set is 83.14%, the loss is 0.049, and the IoU of a single cabbage plant is 82.7%;

[0074] According to the results of two-dimensional image segmentation, the width, spread and reflectivity of different spectral channels of each cabbage were calculated using pixel points combined with ground sampling intervals. The height of the cabbage was calculated based on the segmentation results of the three-dimensional point cloud, and the elevation result of the cabbage based on DSM was calculated based on the segmentation results of DSM (the DSM maximum value of each cabbage minus the ridge height value is the height of a single cabbage). The actual size of the marker was used to calculate the true height combined with the conversion value of each point in the point cloud. At the same time, sphere fitting was performed based on the segmented point cloud of the single cabbage, and ellipse fitting of the canopy was performed based on the plant width to calculate the spread of the cabbage.

[0075] S42. Calculate the length and width pixels of a single Chinese cabbage plant based on the 2D image segmentation results, combine the ground sampling distance GSD, use the ground manual measurement plant spread data as data correction, and calibrate the prediction model. The optimal prediction model accuracy of plant length and width is R 2 They are 0.85 and 0.82 respectively;

[0076] S43. Based on the multispectral image and vegetation index image of a single Chinese cabbage, a prediction model for chlorophyll and nitrogen content was established. This part used ground-measured nitrogen content and real chlorophyll data as corrections. The accuracy of the optimal prediction model for chlorophyll and nitrogen content was R 2 They are 0.75 and 0.72 respectively.

[0077] S44. Based on the 3D point cloud segmentation results, the length, width and height of the Chinese cabbage were calculated and combined with the actual marker size. The plant height data measured manually on the ground were used as data correction. The model prediction accuracy of plant length, plant width and plant height were 0.79, 0.70 and 0.67 respectively, and the prediction accuracy of plant height based on DSM was 0.73. The results of Chinese cabbage plant phenotype based on 3D point cloud computing are shown in Figure 4. Figure 4 As shown, Figure 4 a and Figure 4 b is the horizontal and vertical extension of the Chinese cabbage. Figure 4 c and Figure 4 d are the height prediction results of Chinese cabbage based on 3D point cloud numerical calculation and DSM elevation calculation, respectively. Figure 4 c and 4d clearly show the comparison between the prediction accuracy based on point cloud and elevation.

[0078] S45. Based on the dual-input multimodal convolutional neural network model, a prediction model for chlorophyll and nitrogen content was established.

[0079] Wherein, in step S4, a leaf color quantitative acquisition model is constructed, including:

[0080] Based on the multispectral image of a single Chinese cabbage plant, a leaf color quantitative acquisition model was established, and the average reflectance of different bands of a single Chinese cabbage plant was calculated and the spectral values ​​were combined to obtain different characteristic variables.

[0081] According to different characteristic variables and collected images of single Chinese cabbage, a dual-input convolutional neural network model was established for Chinese cabbage growth monitoring.

[0082] Construct a quantitative acquisition model for the expansion degree of Chinese cabbage, including:

[0083] The 3D point cloud parameters of a single Chinese cabbage plant are obtained by performing spherical fitting, circular fitting and projection processing on the 3D point cloud of a single Chinese cabbage plant, and different quantitative indicators are obtained by performing ratio combination on the 3D point cloud parameters of a single Chinese cabbage plant.

[0084] The principal component analysis was performed on the different quantitative indicators obtained, and the maximum and minimum normalization was performed on each principal component in combination with the contribution weight of each principal component to obtain the value of the comprehensive quantitative Chinese cabbage canopy expansion.

[0085] The extracted quantitative indicators of Chinese cabbage canopy expansion were subjected to multivariate stepwise regression analysis with the values ​​of comprehensive quantitative Chinese cabbage canopy expansion, a leaf color quantification model was established, and indicators with strong correlation were extracted for the quantification of plant expansion in the later stage.

[0086] In addition, in step S4, a dual-input convolutional neural network model is established, including:

[0087] According to the multispectral images of individual Chinese cabbage plants segmented at the rosette stage, the leaf color quantification value was obtained by combining the reflectance of different bands and the band combination value using the leaf color definition formula in S3.

[0088] Based on different characteristic variables and the collected images of a single cabbage plant, a dual-input convolutional neural network model was established for cabbage growth monitoring. The model inputs a five-channel image at one end and calculates the combined value of each band and vegetation index at the other end, which can more accurately predict the chlorophyll content of cabbage and the nitrogen content during the growth period. Cabbage has different nitrogen requirements at different growth stages. The prediction of nitrogen content can help farmers achieve precise fertilization, adjust the amount of fertilizer according to the real-time nitrogen content, avoid wasting fertilizer resources, and improve fertilization efficiency.

[0089] Finally, combined with the results in S2, the phenotypic traits of each plant during the growth period can be tracked, and the changes in the cabbage phenotype and nutrients can be quickly and non-destructively obtained and quantified. Such information can provide a scientific basis for breeders, guide breeding work, and accelerate the cultivation and promotion of high-quality cabbage varieties.

[0090] In order to verify the feasibility of the manual measurement of plant width and plant height correction data on the ground in this embodiment, ten cabbages were used for repeated measurement experiments, and Gage R&R was used for data analysis to determine the type and size of errors that occurred during measurement. The mean-range method (XR) was used to achieve the repeatability and reproducibility of the phenotypic determination of Chinese cabbage. The test measured data 300 times, and the measurement results of the phenotypic plant width are shown in Table 2.

[0091] Table 1 Weather conditions during the UAV operation period

[0092]

[0093] Table 2 X-R method data analysis results plant width

[0094]

[0095]

[0096] The measurement results of the phenotypic plant height are shown in Table 3.

[0097] Table 3 X-R method data analysis results plant height

[0098]

[0099] Based on the analysis results of Table 2 and Table 3, from the "Contribution rate of variance components" column, the variance component contribution rates of the total measuring tool R&R, repeatability, and reproducibility of the Chinese cabbage phenotypic classification model system are 0.04%, 0.03%, 0.01% and 0.06%, 0.06%, 0.00%, respectively, which are far less than 1%. The test results show that the classification model system is acceptable; from the "Research variation rate" column, the total measuring tool R&R% is 2 and 2.42%, which is less than 10%, indicating that the classification system is also acceptable, and its repeatability and reproducibility are 1.74%, 0.985, 2.42%, 0.00%, respectively, which are far less than 10%, which is completely within the acceptable range. According to the distinguishable category number series, they are 70 and 58, respectively, which are far greater than 5, indicating that its measurement accuracy is high. Therefore, it can be considered that the measurement results of Chinese cabbage plant width and plant height can be used for ground correction data for later UAV measurement of Chinese cabbage phenotypes.

[0100] Therefore, the above-mentioned high-throughput acquisition and monitoring method of Chinese cabbage phenotype using multi-source data of unmanned aerial vehicles is adopted, based on the combination of unmanned aerial vehicle multispectral imaging and deep learning, to realize the quantitative acquisition of Chinese cabbage phenotype and the growth monitoring of Chinese cabbage throughout the growth period, with short sampling time, and can accurately predict the phenotypic traits of Chinese cabbage throughout the growth period, replacing time-consuming and labor-intensive methods such as manual visual inspection and measurement, with the characteristics of fastness and accuracy. The multispectral image data obtained by unmanned aerial vehicle technology can provide objective and quantitative indicators for the screening and evaluation of genotypes, help to identify genotypes with excellent phenotypic traits, and provide valuable reference for the rapid acquisition of phenotypic data for the cultivation of different genotypes of Chinese cabbage.

[0101] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0102] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles, characterized in that: The following steps are involved: S1, using drones to collect images of multiple plots of different varieties of Chinese cabbage at different growth stages; S2. Generate digital orthophotos, digital surface models and dense three-dimensional point clouds of different varieties of Chinese cabbage by image stitching based on the collected images of different growth stages; S3, combining the acquired digital orthophoto, digital surface model and dense three-dimensional point cloud with deep learning to segment individual Chinese cabbage plants, and obtain two-dimensional and three-dimensional phenotypic information of individual Chinese cabbage plants; S4. Construct a prediction model for chlorophyll and nitrogen content, a quantitative acquisition model for leaf color, and a quantitative acquisition model for Chinese cabbage expansion based on a dual-input convolutional neural network model to monitor the growth of Chinese cabbage during its growth period; Construct a prediction model for chlorophyll and nitrogen content based on a dual-input convolutional neural network model, including: S41. The Chinese cabbage in the seedling stage was identified and located using the YOLOv8 transfer learning model. The position was calibrated using QGIS software combined with the target detection results. The GPS position of each plant recorded by RTK was found using each center point. S42. Use the improved UNet semantic segmentation model combined with the watershed algorithm to segment individual cabbages, perform individual segmentation on each cabbage in the RGB image and multispectral image, and obtain the two-dimensional image segmentation result. Use the improved PointNet++ point cloud semantic segmentation model to perform complex background removal operations on the sub-regions of the three-dimensional point cloud, and combine with the density clustering algorithm to achieve individual cabbage segmentation, and obtain the three-dimensional point cloud segmentation result of the individual plant. S43, calculating the length and width pixel points of a single Chinese cabbage plant, a vegetation index image, and a multispectral image according to the two-dimensional image segmentation result; S44, calculating the length, width and height of the bounding box of a single Chinese cabbage plant according to the segmentation result of the three-dimensional point cloud, and extracting an index for quantifying the spread of the Chinese cabbage canopy; S45. Based on the dual-input multimodal convolutional neural network model, a prediction model for chlorophyll and nitrogen content was established; Construct a leaf color quantitative acquisition model, including: Based on the multispectral image of a single Chinese cabbage plant, a leaf color quantitative acquisition model was established, and the average reflectance of different bands of a single Chinese cabbage plant was calculated and the spectral values ​​were combined to obtain different characteristic variables. According to different characteristic variables and the collected images of individual Chinese cabbage plants, a dual-input convolutional neural network model was established for Chinese cabbage growth monitoring. Construct a quantitative acquisition model for the expansion degree of Chinese cabbage, including: The 3D point cloud parameters of a single Chinese cabbage plant are obtained by performing spherical fitting, circular fitting and projection processing on the 3D point cloud of a single Chinese cabbage plant, and different quantitative indicators are obtained by performing ratio combination on the 3D point cloud parameters of a single Chinese cabbage plant. The principal component analysis was performed on the different quantitative indicators obtained, and the maximum and minimum normalization was performed on each principal component in combination with the contribution weight of each principal component to obtain the value of the comprehensive quantitative Chinese cabbage canopy expansion. The extracted quantitative indicators of Chinese cabbage canopy expansion were subjected to multivariate stepwise regression analysis with the values ​​of comprehensive quantitative Chinese cabbage canopy expansion, a leaf color quantification model was established, and indicators with strong correlation were extracted for the quantification of plant expansion in the later stage.

2. The method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles according to claim 1, characterized in that: In step S1, the drone-based acquisition of multiple images of different varieties of Chinese cabbage at different growth stages includes: A drone equipped with a five-channel multispectral camera was used to obtain images of different varieties of Chinese cabbage at different growth stages, and multi-angle circular cross-route shooting was used to obtain image data from different angles.

3. The method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles according to claim 1, characterized in that: In step S2, the image stitching is generated in the following way: The collected images at different growth stages are stitched together through UAV image synthesis to obtain a digital orthophoto of the entire field, and a dense three-dimensional point cloud and digital surface model of the field are generated through a registration algorithm.

4. The method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles according to claim 1, characterized in that: In step S3, segmenting a single Chinese cabbage plant includes: The images provided by the digital orthophotos were segmented into 640×640 sub-images through the repetition rate segmentation program, and individual cabbage plants were marked using Labelme for the subsequent training of the UNet model. The collected images were segmented into multiple sub-regions using ClouCompare, and the point cloud of individual cabbage plants was marked for the subsequent training of the PointNet++ model. The phenotypic data of individual cabbage plants were obtained through two-dimensional and three-dimensional information.

5. The method for high-throughput acquisition and monitoring of Chinese cabbage phenotypes using multi-source data from unmanned aerial vehicles according to claim 4 is characterized in that: The two-dimensional and three-dimensional phenotypic information of a single Chinese cabbage plant includes plant height, plant width, spread and leaf color.

Images