Evaluation method for seedling emergence uniformity of mechanically-sown corn

Abstract

The invention discloses a method for evaluating seedling emergence uniformity of mechanically-sown corn. A field high-definition RGB or multispectral image is obtained by using an unmanned aerial vehicle remote sensing platform, and a high-precision three-dimensional point cloud is automatically generated through an SfM-MVS algorithm, so that the problems of high subjectivity and low efficiency of traditional manual sampling are effectively avoided; secondly, automatically extracting a canopy height model and a point density distribution map based on the point cloud data, respectively calculating a height entropy and a density entropy, and quantifying the spatial variability of the crop population from two dimensions of a vertical structure and horizontal distribution; and then, the height entropy and the density entropy are fused to construct a combined entropy uniformity index (CEU), and multi-dimensional and automatic comprehensive evaluation of the emergence uniformity is realized. The invention aims to provide a high-throughput and high-precision technical means for mechanical seeding quality evaluation, crop population consistency monitoring and precision agricultural management.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of analyzing materials by optical means, and particularly relates to a method for evaluating the uniformity of corn emergence after mechanized sowing. BACKGROUND

[0002] Corn is one of the important crops in the northern region. In recent years, with the improvement of agricultural mechanization level and the continuous increase of corn planting area, the sowing quality directly affects the spatial distribution of crop population, light energy utilization efficiency and final yield.

[0003] Traditional crop uniformity detection mainly relies on manual measurement and evaluation. During implementation, the detection personnel need to measure the row spacing, plant spacing and plant height row by row, which not only has high labor intensity and low detection efficiency, but also has a certain degree of deviation in eye recognition, so the detection accuracy is not high, and it is difficult to realize complete detection on a large field. The computer vision measurement method based on visible light image introduces machine vision, which has the advantages of high efficiency and high precision, but still has the following shortcomings: first, the acquisition and processing of field images are easily disturbed by complex environments such as light and straw covering; second, there is a lack of comprehensive uniformity evaluation index integrating multi-dimensional information such as spatial distribution, plant height and density; third, the existing methods have poor generalization ability under different crop types and different operation speeds, and it is difficult to fully reflect the actual operation effect of the seeding machine.

[0004] Therefore, there is an urgent need for an efficient, objective and generalizable technical method that can comprehensively evaluate the sowing quality and emergence uniformity. SUMMARY

[0005] In view of the problems in the background art, the application provides a method for evaluating the emergence uniformity of corn after mechanized sowing, which can realize high-throughput and high-efficiency sowing quality evaluation, and has the advantages of simple operation, objective results and wide applicability. It can be popularized to crops that are sown or planted in discrete rows and have obvious plant row structure. The technical scheme comprises the following steps:

[0006] Step 1: acquiring image data of the target field by a drone carrying a camera;

[0007] Step 2: generating point cloud data based on the motion recovery structure-multiview stereo vision method;

[0008] Step 3: constructing a canopy height model and a point density map and calculating height entropy and density entropy according to the model and the map;

[0009] Step 4: fusing the height entropy and the density entropy to construct a combined entropy uniformity index and comprehensively evaluating the spatial distribution and growth consistency of crops, wherein:

[0010] Combination entropy uniformity

[0011] In the formula, the weight coefficients a, b are obtained by cross-validation and variance minimization criteria; through five-fold cross-validation, a, b ∈ [0, 1] are traversed with a step of 0.1, and the weight pair with the minimum CEU variance is selected, that is, a, b are determined; is the height entropy, is the density entropy;

[0012] Step 5, obtaining the evaluation result of the uniformity of emergence of the target plot according to the combined entropy value of the corn seedlings.

[0013] The flight height of the unmanned aerial vehicle carrying the camera in step 1 is 15 meters, and the overlap rate of the heading and the lateral is not less than 75%.

[0014] After obtaining the image data in step 1, orthorectification and radiation correction are performed, and the geometric accuracy is improved by using ground control points.

[0015] Step 2 includes:

[0016] Step 2.1, using Agisoft Metashape or similar software to perform motion recovery structure and multi-view stereo processing to generate high-density three-dimensional point cloud data excluding ground points;

[0017] Step 2.2, manually or automatically cropping the operation area sub-point cloud corresponding to each seeding machine; preferably, manually cropping the operation area sub-point cloud corresponding to each seeding machine, as shown in Figure 2 .

[0018] Step 3 includes the following steps:

[0019] Step 3.1, for each sub-point cloud, ground point elimination and noise removal based on the RANSAC algorithm;

[0020] Step 3.2, dividing the XY plane of each sub-point cloud into 2cm×2cm grids;

[0021] Step 3.3, for each grid, counting the maximum height value of the plant point cloud in the grid to construct a canopy height model, thereby obtaining a two-dimensional matrix H, D depicting the vertical structure and horizontal density of the target plot, respectively; counting the number of points in the grid to construct a point density map.

[0022] Step 3.3 includes the following steps:

[0023] Step 3.3.1, obtaining the height entropy according to the canopy height model of the plants in the target plot;

[0024] According to the canopy height model (Canopy Height Model, CHM) of the plants in the target plot, the maximum height value distribution of each grid can be obtained, denoted as , wherein N is the total number of grids. The height value distribution is discretized into m intervals, and the first probability of each interval is calculated :

[0025] （1）

[0026] wherein, N k is the number of grids falling into the kth height interval, and m is the total number of intervals. The height entropy is calculated according to :

[0027] （2）

[0028] wherein, H represents the unevenness of the canopy height distribution, and the higher the value, the greater the difference in plant height and the worse the vertical structure consistency of the field; the higher the height entropy H , the greater the difference in plant height, and the more likely to occur lodging or environmental stress; otherwise, the smaller the value, the more uniform the height.

[0029] Step 3.3.2, according to the point density map of the plants in the target field, the density entropy is obtained, and the density entropy is

[0030] Under the same grid division, the number of plant point clouds in each grid is counted, denoted as N k , and a density histogram is constructed; the second probability of the kth density interval is defined as :

[0031] （3）

[0032] wherein, N k is the number of grids falling into the kth density interval; and the formula for calculating the density entropy is

[0033] （4）

[0034] wherein, represents the uniformity of the horizontal distribution of the point cloud, and the higher the value, the more significant the difference in the distribution of plants in the field, and there are phenomena such as missing planting or heavy plants leading to uneven density; otherwise, it indicates that the distribution is uniform.

[0035] The method for obtaining the evaluation result in step 5 includes the following steps:

[0036] The evaluation grade of the uniformity of the emergence of the mechanized sown corn is:

[0037] When CEU≥5.85, the evaluation grade is "very uneven";

[0038] When 5.20≤CEU<5.85, the evaluation grade is "generally uniform";

[0039] When 4.76 ≤ CEU < 5.20, the evaluation level is "uniform";

[0040] When CEU < 4.76, the evaluation level is "very uniform".

[0041] The beneficial effects of this invention are as follows:

[0042] 1. By integrating the height information and two-dimensional density information of three-dimensional point clouds, a combined entropy uniformity index (CEU) was constructed, which can comprehensively evaluate seedling uniformity from both vertical structure and horizontal distribution dimensions, making it more comprehensive and accurate than a single index.

[0043] 2. The use of UAV remote sensing technology enables non-destructive, large-scale, and high-throughput data collection, overcoming the shortcomings of traditional manual surveys, such as low efficiency, strong subjectivity, and limited sampling.

[0044] 3. It is easy to operate and the results are less affected by subjective factors; moreover, the evaluation level is objectively determined based on the magnitude of the evaluation value, making it more acceptable. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating an embodiment of a method for evaluating the uniformity of corn emergence during mechanized sowing according to the present invention.

[0046] Figure 2 This refers to the point cloud of the target field plots manually segmented in this embodiment of the invention.

[0047] Figure 3 This is a schematic diagram of a canopy height model with a resolution of 2cm in an embodiment of the present invention;

[0048] Figure 4 This is a point density distribution map divided into 2cm grids in an embodiment of the present invention;

[0049] Figure 5 This is a frequency histogram of canopy height distribution in an embodiment of the present invention. Detailed Implementation

[0050] The present invention will be further described in detail below with reference to the accompanying drawings.

[0051] like Figure 1 The embodiment of the present invention shown includes the following steps:

[0052] Step 1: Acquire image data of the target field using a drone equipped with a camera;

[0053] In this embodiment, the drone equipped with the camera in step 1 flies at an altitude of 15 meters, and the overlap rate in both the heading and lateral directions is not less than 75%. After acquiring the image data, orthorectification and radiometric correction need to be performed, and ground control points are used to improve geometric accuracy. The image data is RGB or multispectral images.

[0054] Step 2, the point cloud data generated based on the structure from motion-multiple view stereo method (SfM-MVS algorithm) includes:

[0055] Step 2.1, using Agisoft Metashape or similar software to perform structure from motion (SfM) and multiple view stereo (MVS) processing to generate high-density three-dimensional point cloud data with ground points removed;

[0056] Step 2.2, manually or automatically cropping out the operation area sub-point cloud corresponding to each seeding machine; preferably, manually cropping out the operation area sub-point cloud corresponding to each seeding machine, as shown in Figure 2 .

[0057] Step 3, constructing a canopy height model and a point density map and calculating height entropy and density entropy based thereon, including the following steps:

[0058] Step 3.1, for each sub-point cloud, removing ground points and noise points based on the RANSAC algorithm;

[0059] Step 3.2, dividing the XY plane of each sub-point cloud into 2cm×2cm grids;

[0060] Step 3.3, for each grid, counting the maximum height value of the plant point cloud inside the grid to construct a canopy height model, as shown in Figure 3 , thereby obtaining a two-dimensional matrix H, D depicting the vertical structure and horizontal density of the target field, respectively; counting the number of points in the grid to construct a point density map as shown in Figure 4 .

[0061] Specifically, the Shannon entropy formula is introduced to calculate the height entropy and the density entropy, and step 3.3 includes the following steps:

[0062] Step 3.3.1, obtaining height entropy according to the canopy height model of the plants in the target field;

[0063] According to the canopy height model (CHM) of the plants in the target field, the maximum height value distribution of each grid can be obtained, denoted as , where , N is the total number of grids. Discretize the height value distribution into m intervals, and calculate the first probability of each interval:

[0064] （1）

[0065] where, is the number of grids falling into the kth height interval, and m is the total number of intervals. According to this, the height entropy is calculated :

[0066] （2）

[0067] wherein, characterizes the unevenness of the canopy height distribution, the higher the value, the greater the difference in plant height, the worse the vertical structure consistency of the field; height entropy the higher the value, the greater the difference in plant height, and the more likely to occur lodging or environmental stress; otherwise, the smaller the value, the more uniform the height

[0068] Step 3.3.2, according to the point density map of the plants in the target field, the density entropy is obtained, and the density entropy is:

[0069] Under the same grid division, the number of plant point clouds in each grid is counted, denoted as , and a density histogram is constructed; the second probability of the kth density interval is defined as :

[0070] （3）

[0071] wherein, is the number of grids falling into the kth density interval; accordingly, the formula for calculating the density entropy is:

[0072] （4）

[0073] wherein, characterizes the uniformity of the horizontal distribution of the point cloud, the higher the value, the more significant the difference in the distribution of plants in the field, and there are phenomena such as missing planting or heavy plants leading to uneven density; otherwise, it indicates that the distribution is uniform.

[0074] Step 4, fuse the height entropy and the density entropy, construct the combined entropy uniformity index, and comprehensively evaluate the spatial distribution and growth consistency of crops, specifically using the combined entropy uniformity calculation formula to calculate the uniformity of the plant spacing:

[0075] Combination entropy uniformity （5）

[0076] The weight coefficients a, b are obtained through cross-validation and variance minimization criteria; through five-fold cross-validation, a, b ∈ [0, 1] is traversed with a step of 0.1, and the weight pair with the minimum CEU variance is selected to determine a, b; is the height entropy, is the density entropy;

[0077] Step 5, according to the combined entropy value of the corn seedling group, the evaluation result of the uniformity of seedling emergence of the target field is obtained, and the method for obtaining the evaluation result includes the following steps:

[0078] The evaluation grade of the uniformity of corn seedling emergence of mechanical sowing is:

[0079] When CEU≥5.85, the granularity condition is extremely uneven, and the point cloud distribution shows obvious agglomeration or broken band characteristics, which is judged as missing planting or serious over-planting of the plant, and the evaluation level is very "uneven";

[0080] When 5.20≤CEU<5.85, the granularity condition is uneven, and there are obvious gaps or agglomerations in local blocks, which is judged as uneven distribution of seedling and poor stability of seed metering performance of the machine, and the evaluation level is "general uniformity";

[0081] When 4.76≤CEU<5.20, the granularity condition is relatively uniform, and there is slight fluctuation in local areas, which is judged as good seeding quality and stable spatial distribution of plants, and the evaluation level is "uniform";

[0082] When CEU<4.76, the granularity condition is clear and uniform, and the point cloud distribution has good continuity, which is judged as high consistency of seedling and stable seeding performance of the machine, and the evaluation level is "very uniform".

[0083] In specific implementation, the experimental site is located in the test base of LiZong Agricultural Machinery Farmers' Professional Cooperative in Huantai County, Shandong Province. The sown corn variety is 'Huangjinliang MY73'.

[0084] The sowing method is as follows: the previous crop in the test site is wheat, and the wheat is harvested by a wheat combine harvester. Twenty corn sowing plots are delineated for corn sowing, and 20 different types of precision seeders are selected. The row spacing is set to 40 cm and 60 cm wide-narrow row planting, and the theoretical plant spacing is 18 cm.

[0085] The process of evaluating the uniformity of seedling is as follows:

[0086] Step 1, after the corn is sown, the flight height of the unmanned aerial vehicle carrying the camera is set to 15 meters, and the lateral overlap rate is not less than 75%. In this embodiment, the unmanned aerial vehicle platform is DJI Phantom 4 Multispectral, which carries a multispectral camera and can simultaneously collect blue, green, red, red edge and near-infrared band images. In this embodiment, after obtaining the image data, orthographic correction and radiation correction are required, and ground control points are used to improve the geometric accuracy.

[0087] Step 2, in this embodiment, Agisoft Metashape is used for structure from motion (SfM) and multi-view stereo (MVS) processing to generate high-density three-dimensional point cloud data, and the operation area sub-point cloud corresponding to each seeding machine is manually cropped.

[0088] Step 3, for each sub-point cloud, the ground points are removed based on the RANSAC algorithm and the noise points are removed; the XY plane of each sub-point cloud is divided into 2cm×2cm grids; for each grid The plant point cloud statistics point cloud height of the ground point removal inside , and the number of points ; thereby obtaining a two-dimensional matrix H, D respectively depicting the vertical structure and horizontal density of the target field block inside; obtaining height entropy according to the canopy height model of the plants inside the target field block; obtaining density entropy according to the point density map of the plants inside the target field block; the higher the height entropy , the greater the plant height difference, and the more likely to occur lodging or environmental stress; otherwise, the smaller the value, the more uniform the height. The higher the density entropy H d , the poorer the horizontal spatial uniformity, and the more likely to occur uneven density caused by more missing planting / double plants, etc.

[0089] Step 4, calculate the uniformity of plant spacing by using the combined entropy uniformity calculation formula:

[0090] （2）

[0091] Wherein the weight coefficients α, β are obtained by cross-validation and variance minimization criterion; through five-fold cross-validation, traverse α, β ∈ [0, 1] with 0.1 step, and select the weight pair with the minimum CEU variance to determine α, β;

[0092] Step 5, according to the calculated CEU values of 20 target field blocks, obtain the evaluation grade of corn emergence uniformity of the target field block;

[0093] Specifically, the evaluation grade of the emergence uniformity of the mechanized sown corn is:

[0094] When CEU ≥ 5.85, the granularity is extremely uneven, and the point cloud distribution shows obvious agglomeration or broken band characteristics, which is judged as missing planting or serious overplanting of plants, and the evaluation grade is very "uneven";

[0095] When 5.20 ≤ CEU < 5.85, the granularity is uneven, and there are obvious gaps or agglomerations in local blocks, which is judged as uneven distribution of emergence and poor stability of the seeding performance of the machine, and the evaluation grade is "generally uniform";

[0096] When 4.76 ≤ CEU < 5.20, the granularity is relatively uniform, and there is slight fluctuation in local areas, which is judged as good seeding quality and stable spatial distribution of plants, and the evaluation grade is "uniform";

[0097] When CEU < 4.76, the granularity is clear and uniform, and the point cloud distribution has good continuity, which is judged as high consistency of emergence and stable seeding performance of the machine, and the evaluation grade is "very uniform".

[0098] It is calculated that the CEU value of a target plot in the plot is 4.2136, which is determined as "very uniform"; and the CEU value of another target plot is 4.9909, which is determined as "uniform".

[0099] The effective large-scale monitoring of the uniformity of crop growth based on the mechanical seeding of farmland plots is realized, the technical scheme provided by the application fully utilizes the characteristics that the unmanned aerial vehicle remote sensing data can be obtained in a large range, high throughput and lossless, and the crop uniformity information is regularly and discretely distributed, and the crop growth uniformity of the field planting plot is evaluated, the disadvantages of time-consuming and laborious and low efficiency in the previous crop growth uniformity investigation are overcome, the work efficiency is improved, the work intensity is reduced, and the accuracy and precision of the large-scale crop growth uniformity monitoring are effectively improved.

[0100] Compared with the prior art, the unmanned aerial vehicle multi-view stereo reconstruction technology is introduced in the data acquisition layer to generate high-precision point clouds, and an information fusion framework based on entropy theory is established in the evaluation logic. By jointly modeling the canopy height distribution and the point cloud density distribution, the vertical consistency and the horizontal uniformity of the plant spatial structure are first expressed in the same index system, and the multi-dimensional information quantization of the seeding quality is realized. Further, by introducing the cross-validation and variance minimization criteria to adaptively determine the weight parameters α and β, the parameter self-learning optimization is realized in the index construction, the subjectivity and instability of the traditional experience weight setting are overcome, and thus a set of transplanting uniformity evaluation logic which is transferable and generalizable is formed.

Claims

1. A method for evaluating the uniformity of corn emergence during mechanized sowing, characterized in that, include: Step 1: Acquire image data of the target field using a drone equipped with a camera; Step 2: Point cloud data generated based on the structure-of-motion motion reconstruction method (SORP) and multi-view stereo vision method; Step 3: Construct a canopy height model and a point density map, and calculate the height entropy and density entropy accordingly; Step 4: Integrate height entropy and density entropy to construct a combined entropy uniformity index, which comprehensively evaluates the spatial distribution and growth consistency of crops, including: In the formula, the weight coefficients 𝛼,𝛽 are obtained through cross-validation and the variance minimization criterion; through five-fold cross-validation, 𝛼,𝛽∈[0,1] are traversed with a step size of 0.1, and the weight pair with the smallest CEU variance is selected to determine 𝛼,𝛽; For high entropy, Density entropy; Step 5: Obtain the evaluation result of the seedling uniformity of the target plot based on the corn seedling combination entropy value.

2. The method for evaluating the uniformity of corn emergence during mechanized sowing according to claim 1, characterized in that, In step 1, the drone equipped with the camera flies at an altitude of 15 meters, and the overlap rate of the heading and lateral directions is not less than 75%.

3. The method for evaluating the uniformity of corn emergence during mechanized sowing according to claim 1, characterized in that, Step 1 involves performing orthorectification and radiometric correction after acquiring image data, and using ground control points to improve geometric accuracy.

4. The method for evaluating the uniformity of corn emergence during mechanized sowing according to claim 1, characterized in that, Step 2 includes: Step 2.1: Use Agisoft Metashape or similar software to perform motion reconstruction and multi-view stereo processing to generate high-density 3D point cloud data with ground points removed; Step 2.2: Manually or automatically crop out the sub-point cloud of the working area corresponding to each seeding machine; preferably, manually crop out the sub-point cloud of the working area corresponding to each seeding machine, as shown in Figure 2.

5. The method for evaluating the uniformity of corn emergence during mechanized sowing according to claim 1, characterized in that, Step 3 includes the following steps: Step 3.1: For each sub-point cloud, perform ground point culling and noise removal based on the RANSAC algorithm; Step 3.2: Divide the XY plane of each sub-point cloud into a 2cm×2cm grid; Step 3.3: For each grid cell, count the maximum height of the plant point cloud within it (excluding the location surface) to construct a canopy height model. This yields two-dimensional matrices H and D, which respectively characterize the vertical structure and horizontal density within the target field. Count the number of points within the grid cell to construct a point density map.

6. The method for evaluating the uniformity of corn emergence during mechanized sowing according to claim 5, characterized in that, Step 3.3 includes the following steps: Step 3.3.1: Obtain the height entropy based on the canopy height model of the plants in the target field; The distribution of maximum height values ​​for each grid cell can be obtained based on the canopy height model (CHM) of the plants within the target field, denoted as . ,in N is the total number of grid cells. The height value distribution is discretized into m intervals, and the first probability of each interval is calculated. : in, Let m be the number of grid cells falling into the k-th height interval, and m be the total number of intervals. Calculate the height entropy based on this. : In the formula, The height entropy characterizes the degree of unevenness in canopy height distribution; a higher value indicates greater differences in plant height and poorer uniformity in the vertical structure of the field. A higher value indicates greater variation in plant height, which may lead to lodging or environmental stress; conversely, a lower value indicates more uniform plant height. Step 3.3.2: Obtain the density entropy based on the point density map of plants in the target field. Density entropy: Under the same grid division, count the number of plant point clouds in each cell, and record it as . And construct a density histogram; define the second probability of the k-th density interval #imgpt13#: Where #imgpt15# represents the number of grid cells falling into the k-th density interval; the formula for calculating the density entropy is as follows: In the formula, #imgpt17# represents the uniformity of the horizontal distribution of the point cloud. The higher the value, the more significant the differences in the distribution of plants in the field, and the existence of phenomena such as missed sowing or double sowing leading to uneven density; conversely, it indicates that the distribution is uniform.

7. The method for evaluating the uniformity of corn emergence during mechanized sowing according to claim 1, characterized in that, The method for obtaining the evaluation results in step 5 includes the following steps: The evaluation level for uniform emergence of corn seedlings in mechanized sowing is as follows: When CEU ≥ 5.85, the evaluation level is "very uneven"; When 5.20≤CEU<5.85, the evaluation level is "average and uniform"; When 4.76 ≤ CEU < 5.20, the evaluation level is "uniform"; When CEU < 4.76, the evaluation level is "very uniform".

Images