A method, device, equipment and medium for estimating field straw covering amount

Abstract

The application discloses a kind of estimation methods, devices, equipment and medium of field straw covering quantity, it is related to farmland conservation tillage technical field, comprising: according to the corresponding relationship of different spectral bands and vegetation index, the index value of different spectral bands corresponding to each sampling point is obtained, the correlation of the index value of different spectral bands corresponding to each sampling point and the straw covering quantity of corresponding sampling point is determined;The straw covering quantity of each sampling point is sorted from small to large, and the sampling point with strong correlation is selected and the index value of corresponding different spectral bands is used to train random forest model;The index value of different spectral bands corresponding to each sampling point in the field research area is input into the model for estimating straw covering quantity to obtain the straw covering quantity of each sampling point;And the average of the straw covering quantity of each sampling point is used as the field straw covering quantity.The application can accurately estimate the straw covering quantity of whole farmland.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of conservation tillage technology in farmland, and particularly relates to a method and device for estimating the amount of straw mulch in a field, equipment and a medium. BACKGROUND

[0002] Straw is the stem and leaf or ear part of mature crops, which contains a large amount of organic matter and is an important source of organic fertilizer in agricultural production. Crop straw mulching and no-till or reduced-till seeding is a major part of modern tillage technology, and the amount and coverage rate of straw mulch are important indicators for measuring the level of straw mulch, and also reflect the quality and effect of straw mulching.

[0003] In the prior art, high-resolution image data of samples in a field research area is obtained by remote sensing technology, and the amount of straw mulch of the samples is estimated using an inversion model in combination with relevant characteristics such as spectral indices.

[0004] However, remote sensing technology can only obtain partial samples for local estimation of the amount of straw mulch, and cannot accurately estimate the amount of straw mulch in the entire field. SUMMARY

[0005] The embodiments of the present application provide a method and device for estimating the amount of straw mulch in a field, equipment and a medium, which can solve the problem in the prior art that the amount of straw mulch in the entire field cannot be accurately estimated.

[0006] The embodiments of the present application provide a method for estimating the amount of straw mulch in a field, which comprises the following steps: obtaining a multispectral remote sensing image of each sampling point in a field research area, and obtaining the amount of straw mulch of each sampling point; extracting different spectral bands in the multispectral remote sensing image of each sampling point, and obtaining index values of different spectral bands corresponding to each sampling point according to the corresponding relationship between different spectral bands and vegetation indices; determining the correlation between the index values of different spectral bands corresponding to each sampling point and the amount of straw mulch of the corresponding sampling point; sorting the amount of straw mulch of each sampling point from small to large, and selecting sampling points with strong correlation; using a random forest model to train and learn the correlation between the index values of different spectral bands and the amount of straw mulch, and using a genetic-chaotic particle swarm hybrid algorithm to optimize the random forest model, to obtain a model for estimating the amount of straw mulch; inputting the index values of different spectral bands corresponding to each sampling point in the field research area into the model for estimating the amount of straw mulch, obtaining the amount of straw mulch of each sampling point, and taking the average value of the amount of straw mulch of each sampling point as the amount of straw mulch in the field.

[0007] Further, the index value of the different spectral band corresponding to each sampling point is obtained by the following steps: using a remote sensing image processing tool to perform band registration processing and geometric correction preprocessing on the multispectral remote sensing image of each sampling point; importing the preprocessed multispectral remote sensing image of each sampling point into a remote sensing image processing software to extract the spectral reflectance data of different spectral bands of each sampling point; and using a vegetation index solving method corresponding to different spectral bands to obtain the index value of the different spectral band corresponding to each sampling point according to the spectral reflectance data of different spectral bands of each sampling point.

[0008] Further, the model for estimating the straw coverage is obtained by the following steps: sorting the straw coverage of each sampling point from small to large, dividing the straw coverage into a modeling set and a verification set in a ratio of 2:1 and performing equal-interval sampling; selecting sampling points with strong correlation between the index value of the different spectral band corresponding to each sampling point and the straw coverage of the corresponding sampling point; using the index value of the different spectral band corresponding to the selected sampling points to train a random forest model to obtain the model for estimating the straw coverage, and using a genetic-chaotic particle swarm hybrid algorithm to optimize the model for estimating the straw coverage; and using the predicted and measured straw coverage of the sampling points as the verification set to perform precision evaluation by using a determination coefficient and a root mean square error evaluation index.

[0009] Further, the multispectral remote sensing image of each sampling point is obtained by the following steps: dividing a field research area into a plurality of sampling units, randomly selecting 6 sampling points in each unit, and setting each straw coverage sampling point as a square region with a size of 0.5m*0.5m; using a multi-spectral camera carried by a UAV to perform vertical shooting in a low-altitude fixed-point manual mode to obtain the multispectral remote sensing image of each sampling point.

[0010] Further, the average value of the straw coverage of each sampling point is obtained by using a Monte Carlo model.

[0011] The embodiment of the present application provides a device for estimating the straw coverage in a field, which comprises:

[0012] The data acquisition module is configured to acquire a multispectral remote sensing image of each sampling point in a field research area and to acquire a straw coverage of each sampling point. The correlation analysis module is configured to extract different spectral bands in the multispectral remote sensing image of each sampling point, to obtain index values of different spectral bands corresponding to each sampling point according to a corresponding relationship between the different spectral bands and a vegetation index, to determine a correlation between the index values of different spectral bands corresponding to each sampling point and the straw coverage of the corresponding sampling point, to sort the straw coverage of each sampling point from small to large, and to select sampling points with strong correlation. The model construction module is configured to use a random forest model to train and learn the correlation between the index values of different spectral bands and the straw coverage, to optimize the random forest model using a genetic-chaotic particle swarm hybrid algorithm, and to obtain a model for estimating the straw coverage. The coverage estimation module is configured to input the index values of different spectral bands corresponding to each sampling point in the field research area into the model for estimating the straw coverage, to obtain the straw coverage of each sampling point, and to take an average of the straw coverage of each sampling point as the field straw coverage.

[0013] The embodiment of the present application provides an electronic device, including a memory and a processor; the memory stores a computer program; and the processor implements the field straw coverage estimation method when executing the computer program.

[0014] The embodiment of the present application provides a computer readable storage medium, which stores a computer program; and the computer program is executed by a processor to implement the field straw coverage estimation method.

[0015] The embodiment of the present application provides a field straw coverage estimation method, device, equipment and medium, and has the following beneficial effects compared with the prior art.

[0016] The correlation analysis module is configured to extract different spectral bands in the multispectral remote sensing image of each sampling point, to obtain index values of different spectral bands corresponding to each sampling point according to a corresponding relationship between the different spectral bands and a vegetation index, to determine a correlation between the index values of different spectral bands corresponding to each sampling point and the straw coverage of the corresponding sampling point, to sort the straw coverage of each sampling point from small to large, and to select sampling points with strong correlation. The model construction module is configured to use a random forest model to train and learn the correlation between the index values of different spectral bands and the straw coverage, to optimize the random forest model using a genetic-chaotic particle swarm hybrid algorithm, and to obtain a model for estimating the straw coverage. The coverage estimation module is configured to input the index values of different spectral bands corresponding to each sampling point in the field research area into the model for estimating the straw coverage, to obtain the straw coverage of each sampling point, and to take an average of the straw coverage of each sampling point as the field straw coverage.

[0017] The present application uses a random forest model to train the correlation between the index values of different spectral bands and the straw coverage, and can obtain accurate straw coverage of multiple sampling points according to the trained model for estimating the straw coverage, and then input the index values of different spectral bands corresponding to each sampling point into the model for estimating the straw coverage, take the average value as the field straw coverage, and finally obtain the field straw coverage estimated by the accurate straw coverage of each sampling point, thereby achieving the purpose of accurately estimating the field straw coverage. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 An implementation flowchart of the field straw coverage estimation method provided by the embodiment of the present application is provided.

[0019] Figure 2 A research area overview provided by the embodiment of the present application is provided.

[0020] Figure 3 A sampling and determination flowchart of the field straw coverage estimation method provided by the embodiment of the present application is provided.

[0021] Figure 4 A Monte Carlo estimation method flowchart provided by the embodiment of the present application is provided.

[0022] Figure 5 A scatter plot of the predicted value and the measured value of the verification set of the field straw coverage estimation method provided by the embodiment of the present application (spring of 22 and autumn of 22) is provided.

[0023] Figure 6 A scatter plot of the predicted value and the measured value of the verification set of the field straw coverage estimation method provided by the embodiment of the present application (spring of 23 and autumn of 23) is provided. DETAILED DESCRIPTION

[0024] In order to make the above-mentioned objects, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the following description, a large number of specific details are set forth in order to facilitate a full understanding of the present application. However, the present application 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 scope of the present application, so the present application is not limited to the specific embodiments disclosed below.

[0025] Referring to Figures 1-6 The embodiment of the present application provides a field straw coverage estimation method, which comprises the following steps:

[0026] Step one: Collecting the straw coverage data of the field research area, using the unmanned aerial vehicle to carry the multispectral camera to collect the multispectral remote sensing image of the field research area, setting the sampling points for the full coverage area through visual observation method, obtaining the multispectral image of the sampling points through low-altitude fixed-point manual control mode, and measuring the straw coverage in the sampling points at the same time.

[0027] Specifically, the DJI Matric 300RTK multi-rotor unmanned aerial vehicle of DJI Innovation Technology Co., Ltd. is used to carry the AQ600 PRO multispectral camera of Changguoyuchen Information Technology and Equipment (Qingdao) Co., Ltd. to collect large-scale field images in the mode of high-altitude automatic route, determine the field coverage type of large-scale area image according to visual observation method, and select the straw full coverage type as the research area. According to the area average division of the research area, 6 sampling points are randomly selected in each unit, each straw coverage sampling point is set as a square area of 0.5m x 0.5m, the unmanned aerial vehicle carrying the multispectral camera is used to vertically shoot in the low-altitude fixed-point manual mode, and the multispectral image of the straw coverage sampling point is obtained. After shooting the multispectral image of the sampling point, the straw coverage in the sampling point is measured, and the straw coverage thickness of four corners and center in each sampling point is measured for error verification.

[0028] Step two: Based on the multispectral image of the sampling point obtained by the unmanned aerial vehicle carrying the multispectral camera, the index values of different spectral bands are constructed and correlated with the straw coverage for the estimation of local straw coverage, i.e. the straw coverage of each sampling point.

[0029] Specifically, the image preprocessing includes: performing band registration processing on the obtained initial single-band image by using the Yusense Map software, performing geometric correction preprocessing, introducing the preprocessed unmanned aerial vehicle multi-spectral image into the ENVI5.3 software for further processing, and obtaining spectral reflectance data of different bands. The present application selects 12 spectral indices, which are normalized difference vegetation index (NDVI), difference vegetation index (DVI), enhanced vegetation index (EVI), brightness index (BI), simple ratio index (SR), green-light normalized difference vegetation index (GNDVI), green-light ratio vegetation index (GRVI), straw multiplier index (SMI), red edge ratio index (SRER), near-infrared normalized index (NRNI), normalized difference red edge index (NDRE), and red-light normalized difference index (RNDI). In view of the above, sample data is selected to perform correlation analysis on different spectral index variables and straw coverage.

[0030] Step three: establishing an estimation model of local straw coverage by using a machine learning and intelligent optimization algorithm, and taking the screened index values of different spectral bands as input variables of the estimation model of straw coverage of each sampling point.

[0031] Specifically, the research area samples are sorted and grouped according to the straw coverage from small to large, the equal interval sampling is carried out according to the ratio of 2:1 of the modeling set and the verification set, the random forest (RF) algorithm is used to establish the straw coverage estimation model based on the strong correlation spectral index, and the genetic-hybrid particle swarm optimization (GA-HPSO) algorithm is used for optimization. Combined with the measured data, the sample data is used to test and analyze the local straw coverage estimation model constructed by the GA-HPSO-RF algorithm, the model prediction value and the measured value are fitted by using the verification set, and the accuracy of the model modeling and verification is evaluated by using the coefficient of determination (R 2 ) and the root mean square error (RMSE) two indexes. R 2 The closer to 1, the smaller the RMSE, and the better the model accuracy effect.

[0032] Step four: using the sample results of the local estimation model, combining the Monte Carlo idea, realizing the global field straw coverage estimation, and completing the large-scale rapid estimation and detection of the field plot corn straw coverage.

[0033] Specifically, combining the Monte Carlo idea, using the Gaussian mixture model and the expectation maximization algorithm, the sample results of the local estimation model are used as the solution of the global straw coverage, and the large-scale detection of the field plot corn straw coverage is completed. The resource amount of the corn straw that can be collected is obtained by using the grass valley ratio, and the global field straw coverage estimation is tested and analyzed, and the accuracy and the error are used as evaluation indexes.

[0034] The present application has the following beneficial effects:

[0035] The present application is based on the multi-rotor unmanned aerial vehicle carrying the multi-spectral camera to obtain a large range of remote sensing images, determines the straw coverage type based on the visual observation method, selects the spectral index by measuring the straw coverage data on the ground, establishes the straw coverage estimation model suitable for the sample point, and realizes the estimation of the global straw coverage by using the Monte Carlo method.

[0036] The random forest algorithm optimized by the genetic-hybrid particle swarm optimization algorithm is used to establish the local estimation model of the corn straw coverage. Under the data of different research areas and different time periods, the model has certain generalization ability and accuracy, the and of the model reach 0.77 and 0.72 on average, and the RMSEv is only 0.050, which achieves the best effect.

[0037] The Monte Carlo idea strategy is introduced to complete the global estimation of the straw covering amount in the field. Based on the local sample straw covering amount obtained by the GA-HPSO-RF model, the global straw covering amount estimation is realized by using the Monte Carlo idea, the accuracy is 86.55% on average, and the error is 0.98t on average. It shows that the application has certain effectiveness and applicability in the detection of straw returning in conservation tillage, meets the requirements of large-scale detection.

[0038] The accurate application of straw covering amount estimation realizes the large-scale detection of the straw covering amount in the field, reduces manual operation, improves work efficiency, provides technical support for effective development of conservation tillage, improves resource utilization efficiency, protects the ecological environment, and promotes the sustainable development of agriculture.

[0039] The embodiment of the application provides an estimation device for the straw covering amount in the field, which comprises:

[0040] The data acquisition module is used for acquiring the multispectral remote sensing image of each sampling point in the field research area, and acquiring the straw covering amount of each sampling point. The correlation analysis module is used for extracting different spectral bands in the multispectral remote sensing image of each sampling point, and obtaining the index value of each sampling point corresponding to different spectral bands according to the corresponding relationship between different spectral bands and vegetation indexes. The correlation between the index value of each sampling point corresponding to different spectral bands and the straw covering amount of the corresponding sampling point is determined. The model construction module is used for sorting the straw covering amount of each sampling point from small to large, and selecting the sampling points with strong correlation. The index value of different spectral bands corresponding to the selected sampling points is used to train and learn the correlation between the index value of different spectral bands and the straw covering amount by using the random forest model, and the random forest model is optimized by using the genetic-chaos particle swarm hybrid algorithm to obtain the model for estimating the straw covering amount. The covering amount estimation module is used for inputting the index value of different spectral bands corresponding to each sampling point in the field research area into the model for estimating the straw covering amount, obtaining the straw covering amount of each sampling point, and taking the average value of the straw covering amount of each sampling point as the straw covering amount in the field.

[0041] The embodiment of the application provides an electronic device, which comprises a memory and a processor; the memory stores a computer program, and the processor realizes the steps of the above-mentioned estimation method for the straw covering amount in the field when executing the computer program.

[0042] The embodiment of the application provides a computer readable storage medium, which stores a computer program, and the computer program realizes the steps of the estimation method for the straw covering amount in the field when being executed by a processor.

[0043] A specific implementation is as follows:

[0044] The protective tillage promotion experimental field in Changchun City was taken as the research area, and corn stalks were taken as the research object. The field remote sensing image was obtained by using unmanned aerial vehicle carrying multispectral camera. The sampling points were set in the full coverage area by visual observation method, and the multispectral image was obtained. The coverage data was measured synchronously. The spectral index was constructed, and the correlation analysis was used to select the spectral index with high correlation as the input variable of the model. The genetic-chaotic particle swarm hybrid algorithm was used to optimize the random forest algorithm, and the local estimation model of corn stalk coverage was constructed. The Monte Carlo thought was used to estimate the global straw coverage by local sampling, and the large-scale detection of straw coverage in the field was realized. A new method was provided for the quantitative estimation of straw coverage by unmanned aerial vehicle remote sensing in protective tillage.

[0045] 1. Collect the straw coverage data in the research area, and use the unmanned aerial vehicle carrying multispectral camera to collect the multispectral remote sensing image of the research area. The sampling points are set in the full coverage area by visual observation method, and the multispectral image of the sampling points is obtained by low-altitude fixed-point manual control mode. At the same time, the straw coverage in the sampling points is measured;

[0046] In this embodiment, the protective tillage promotion demonstration experimental field in Changchun City, Jilin Province, is selected as the research area. Changchun City in Jilin Province is located in the middle of Songliao Plain, and belongs to temperate semi-humid continental monsoon climate. The climate is mild, and the four seasons are distinct. The average annual temperature is 5.5℃, the annual precipitation is 558mm, and the average altitude is about 220m. The main planting crops in the research area are rotation and wide-narrow row planting mode, and the main variety is Oudi598 corn hybrid (Jilin Hongxiang Agricultural Group Hongxiang Seed Co., Ltd.). Spring corn is mostly sown in early May and harvested in early October. The ground in the research area is relatively flat, and the terrain is flat. Jilin Province vigorously promotes protective tillage with straw mulching, and the protective tillage mode with full straw coverage also exists widely. At the same time, there are also some traditional tillage modes, which are easy to compare and suitable for corn straw coverage estimation research.

[0047] 1.1. The DJI Matric 300RTK multi-rotor unmanned aerial vehicle carrying AQ600 PRO multispectral camera of Changguanyuchen Information Technology and Equipment (Qingdao) Co., Ltd. was used to collect large-scale field images in high-altitude automatic route mode. According to the visual observation method, the field coverage type of the large-scale area image was determined, and the straw full coverage type was selected as the research area.

[0048] Specifically, the AQ600 PRO sensor collects data in discrete spectral bands: blue 450 nm, green 555 nm, red 660 nm, red edge 720 nm, near-infrared 840 nm, with a pixel resolution of 2048 x 1536. The AQ600 PRO supports on-board real-time radiometric correction, capturing images of a radiometric calibration target (gray scale plate) before each flight, with a uniform spectral response during imaging, which can improve the accuracy and reliability of measurements, while performing relative radiometric calibration, and enabling downlink light correction during flight through DLS, on-board real-time reflectance calculation, and storage in 8-bit JPEG format.

[0049] In order to ensure the quality of the unmanned aerial vehicle multi-spectral remote sensing image observation, the case was carried out in the research base. The unmanned aerial vehicle image acquisition work was carried out between 10:00 and 15:00 on a clear and cloudless day. The high-altitude unmanned aerial vehicle flight mode was used to obtain complete images of the research area plot. The flight route was planned in advance, with a flight speed of 3 m / s, a lateral overlap rate and a heading overlap rate of 80%. The flight height was selected as 40 m after several test flights. Through visual observation method, the straw coverage amount research area of the case was determined as Jilin Province Changchun City Agricultural Machinery Research Institute scientific research base (research area 1) and Jilin Province Yushu City protective tillage promotion experimental field (research area 2). Research area 1 is located in Erdao District of Changchun City (125°29′56″E, 43°52′06″N), which is approximately a quadrilateral, with an average length of about 150 m in south-north direction and an average width of about 150 m in east-west direction, with a total area of about 22500 m 2 ; Research area 2 is located in Dagang village of No. 8 town of Yushu city (126°13′59″E, 45°05′44″N), which is approximately a quadrilateral, with an average length of about 200 m in south-north direction and an average width of about 100 m in east-west direction, with a total area of about 20000 m 2 . The research area is shown in Figure 2 .

[0050] 1.2. According to the average area of the research area, the sampling unit is divided, and 6 sampling points are randomly selected in each unit. Each straw coverage sampling point is set as a square area of 0.5 m x 0.5 m. The unmanned aerial vehicle is used to carry out vertical shooting in low-altitude fixed-point manual mode to obtain multi-spectral images of straw coverage sampling points.

[0051] Specifically, taking research area 1 as an example, the sampling point multi-spectral image acquisition is shown in Figure 3 . In order to ensure the accuracy of the spectral image and the integrity and accuracy of the data, the case selects multiple time periods and multiple locations for collection, and the sampling time is shown in Table 1 sample collection description.

[0052] The multi-spectral image of the corn straw coverage area sampling point was obtained by using the fixed-point manual control mode, and was taken at the right above the sampling point. The height was determined as 12 m through multiple experiments, and the standard white board was calibrated before the task was executed.

[0053] Table 1 sample collection description

[0054]

[0055]

[0056] 1.3. After taking the multi-spectral image of the sampling point, the straw coverage amount in the sampling point was measured, and the straw coverage thickness was measured at five points including four corners and the center in each sampling point for error verification.

[0057] Specifically, the net weight of the straw coverage sample in each sampling point was measured by using an electronic scale (Hengzi KC-833, precision: 0.001 kg). A total of 360 sampling points were obtained in the research area, and the sampling determination process is as shown in Figure 3 .

[0058] 2. Based on the multi-spectral image of the sampling point obtained by the unmanned aerial vehicle carrying the multi-spectral camera, the spectral index was constructed and the correlation analysis with the straw coverage amount was performed for the estimation of the local straw coverage amount;

[0059] Specifically, the image preprocessing includes: the obtained initial single-band image is subjected to band registration and geometric correction processing by using YusenseMap software, and is imported into ENVI5.3 software for further processing. The spectral image corresponding to each sampling point is cut out on the image, the average reflectance spectrum of the straw coverage sample in the region of interest (ROI) is obtained after removing the marked shadow, and the spectral reflectance data of different bands are obtained.

[0060] Specifically, the spectral indices are constructed, and 12 spectral indices are selected in this case, which are normalized difference vegetation index (NDVI), difference vegetation index (DVI), enhanced vegetation index (EVI), brightness index (BI), simple ratio index (SR), green-light normalized difference vegetation index (GNDVI), green-light ratio vegetation index (GRVI), straw multiplier index (SMI), red edge ratio index (SRER), near-infrared normalized index (NRNI), normalized difference red edge index (NDRE), and red-light normalized difference index (RNDI). The calculation formulas of the spectral indices are shown in Table 2.

[0061] Table 2: Spectral indices and calculation formulas

[0062]

[0063] Note: B, G, R, RE and NIR are the spectral reflectance at 450, 555, 660, 720 and 840 nm wavelengths, respectively.

[0064] In order to analyze the relationship between the spectral indices and the straw coverage, and effectively evaluate the straw coverage, the sample data are selected to perform correlation analysis on the spectral indices and the straw coverage, and the Pearson correlation coefficient method is used, and the calculation formula is as follows:

[0065] T = COV (X, Y) ÷ (σX × σY)

[0066] wherein COV (X, Y) is the covariance of X and Y, and σX and σY are the standard deviations of X and Y, respectively. The obtained results are shown in Table 3: Pearson correlation coefficients of spectral index variables and straw coverage.

[0067] Table 3 Pearson correlation coefficients of spectral index variables and straw mulch amount

[0068]

[0069] Note: * indicates weak correlation, ** indicates moderate correlation, and *** indicates strong correlation

[0070] 3. An estimation model of local straw mulch amount is established by machine learning and intelligent optimization algorithm, and the screened spectral index is used as an input variable of the estimation model of local straw mulch amount;

[0071] Specifically, 360 sample data are sequentially grouped by straw mulch amount from large to small in two periods, and combined with the straw mulch thickness of the sampling points at the time of measurement, the sampling points with large errors are deleted, and random sampling is performed according to the ratio of 2:1 between the modeling set and the verification set. The modeling set and the verification set are as shown in Table 4. In order to obtain a local estimation model with good generalization ability and accuracy, a straw mulch amount estimation model of random forest (RF) algorithm is established, and a genetic algorithm-hybrid particle swarm optimization (GA-HPSO) is used for optimization. The model is completed in MATLAB 2018a software, the hardware environment is Windows 10 operating system, Intel(R) Core(TM) i7-8550U CPU@1.80GHz 2.00GHz processor, 8G memory, and the graphics card is NVIDIA GeForce 940MX.

[0072] Table 4 Modeling set and verification set description of the verification area

[0073]

[0074] In the GA-HPSO algorithm, the crossover and mutation operations in the GA algorithm are embedded in the PSO algorithm, and at the same time, the Logistic chaotic model is used in the initialization process. The RF straw mulch amount estimation model is optimized by using the GA-HPSO algorithm, and in this case, the training set cross-validation is used for decision tree number optimization, and after multiple training, the optimal decision tree number is determined to be 100.

[0075] Specifically, the model predicted value and the measured value are fitted by using the verification set, and the accuracy of the model modeling and verification is evaluated by using two indexes of the coefficient of determination (R 2 ) and the root mean square error (RMSE). R 2 ) and the root mean square error (RMSE).2 The closer to 1, the smaller the RMSE, the better the model precision effect.

[0076] 4. Using the sample results of the local estimation model, combining the Monte Carlo idea, realizing the global straw coverage estimation, completing the large-scale rapid estimation and detection of the field plot corn straw coverage.

[0077] Specifically, the straw coverage of the sampling point is obtained through the estimation model, and according to the Monte Carlo idea, the estimation result of the straw coverage of the sampling point is taken as a known sample, the experience distribution is obtained from the known sample, and the Gaussian mixture model is established. When it is superimposed by k Gaussian distributions, the probability density function formula is as follows:

[0078]

[0079] Where, θ is the overall Gaussian model parameter, is 1, π k is the weight, x is the sample, μ is the model expectation, and ∑ is the model variance. The distribution model parameters of each class in the mixed model are solved by using an iterative optimization strategy EM algorithm. Each iteration process is divided into two steps, one is the expectation step (E step), and the other is the maximum step (M step). The key is to obtain the expectation, calculate the following Q function, and the E step is as follows:

[0080] Q(θ,θ i )=E z [log(X,Z|θ)x,θ i ]=∑logp(X,Z|θ)P(Z|X,θ i )

[0081] Where, Z is the hidden variable, i.e. unknown coverage data, θ=(μ k ,∑ k ,π k ) is the initial parameter estimation value of the Gaussian model, is the parameter estimation value of the Gaussian model after iteration i times, and the joint distribution is P(Z|X,θ), given the observation data X and the parameter estimation θ i The conditional probability distribution P(Z|X,θ i ) of the hidden variable Z is solved, and the M step is as follows:

[0082] θ i+1 =argmaxQ(θ,θ i )

[0083] Maximize Q(θ,θ i ), so as to determine the parameter estimation value θ i+1Under the current obtained latent variable Z and observation variable X, a new parameter θ is needed to make the current state have the highest possibility. In this case, random sampling is used to obtain sample unit sample values, and the expected value of the sample unit straw coverage is obtained. The sum of the sample units is taken as the value of the large-scale straw coverage, so as to realize the estimation of the large-scale field corn straw coverage. The Monte Carlo estimation method process is shown in Figure 4 .

[0084] In order to facilitate evaluation and subsequent calculation, the present application needs to obtain accurate field straw collectable resource quantity. The most widely used grass-to-grain ratio method is used to obtain the corn straw collectable resource quantity. The grass-to-grain ratio is the ratio of the unit area straw yield of a certain crop to the grain yield. The sample collection is a whole plant corn straw sample in the research area, which is dried by using a laboratory electric heating air drying oven, and the net mass of corn grain and corn straw is measured respectively.

[0085] Specifically, the grass-to-grain ratio μ i is as follows:

[0086]

[0087] Wherein, M S is the straw mass, M G is the straw grain mass, A S is the straw moisture content, A G is the grain moisture content. 15% is the moisture content when the straw is air dried, and 12.5% is the national standard water impurity rate of grain crops. The formula of the theoretical resource quantity P of corn straw is as follows:

[0088] P = ∑ μ i G i

[0089] Wherein, G i is the corn yield, and the corn straw collectable resource quantity P c is as follows:

[0090] P c = ∑ δ i μ i G i

[0091] Wherein, δ i is the straw collection coefficient, and the calculation process is as follows:

[0092]

[0093] Wherein, L i is the average height of corn, L i,j is the cutting height of mechanically harvested crops, L i，r is the cutting height of manually harvested crops, and J iFor crop mechanization harvesting rate, Z i For crop straw harvesting and transportation loss rate. According to the actual data statistics, the research area grass valley ratio and corn straw collectable resource amount are shown in Table 5.

[0094] Table 5 Research area corn straw resource statistics

[0095]

[0096]

[0097] For this embodiment (Changchun protective tillage promotion experimental point), the above-mentioned one kind of field straw coverage amount estimation method based on unmanned aerial vehicle multispectral remote sensing image is adopted, and the results obtained and analyzed as follows:

[0098] For step 1, high-altitude images are obtained by unmanned aerial vehicle carrying multispectral camera, and the coverage of field plot is better observed by visual observation method, the straw coverage area is determined, and unnecessary detection time and later screening are reduced. For step 2, the data of field corn straw coverage area is obtained, the spectral information of sampling points is extracted to construct spectral index, which can effectively realize the estimation and statistics of ground parameters, thereby providing rich information for application fields such as environmental monitoring, resource management and decision making. As can be seen from Table 2, 12 multispectral index variables and straw coverage amount all show certain correlation. The correlation from large to small is SRER, NDRE, SR, NDVI, GNDVI, DVI, NRNI, EVI, GRVI, RNDI, BI and SMI, wherein the correlation coefficient of SRER is more than 0.55, the correlation coefficients of NDRE and SR are close to 0.45, the correlation coefficients of NDVI, GNDVI and DVI are close to 0.4, the correlation coefficients with straw coverage amount are significant at 0.001 level, showing strong correlation, which indicates that SRER, NDRE, SR, NDVI, GNDVI and DVI can represent straw coverage amount to a certain extent.

[0099] For step 3, SRER, NDRE, SR, NDVI, GNDVI and DVI are selected as input variables, a straw coverage amount estimation model based on GA-HPSO-RF is established, and model precision evaluation is carried out, and the model precision result is shown in Table 6 based on GA-HPSO-RF model estimation result, wherein RMSE v As the evaluation index of the model on the modeling set, RMSE vAs an evaluation metric in the validation set, to comprehensively evaluate and select the optimal inversion model for corn straw mulch, eight sets of data were synthesized, and the data were fully validated in conjunction with actual needs and feasibility to ensure the reliability and stability of the evaluation results. Table 6 shows the estimation results based on the GA-HPSO-RF model, indicating that the eight sets of data yielded… and The results are quite close, indicating that the model has not overfitted while maintaining high accuracy. The average value is 0.77. The average value is 0.72. and All values ​​reached above 0.70, demonstrating a clear advantage. RMSE c The average value is 0.043, RMSE v The average value is 0.050, RMSE c and RMSE v All values ​​are below 0.059, demonstrating strong stability.

[0100] Table 6. Estimation results based on the GA-HPSO-RF model

[0101]

[0102] The predicted values ​​obtained from the four corn stalk cover estimation models on the validation set were compared with the measured values, and the scatter plot is shown below. Figure 5 and Figure 6 As shown in Table 6, the results are within a reasonable range, demonstrating good fitting ability and high accuracy. The estimation results based on the GA-HPSO-RF model and... Figure 5 ,as well as Figure 6 It can be seen that the R2v values ​​for the four periods in study area 1 were 0.71, 0.75, 0.72, and 0.74, respectively, and the RMSE was [missing value]. v The values ​​were 0.059, 0.045, 0.057, and 0.048, respectively, indicating that the results obtained from the autumn data over the two years were better than those from the spring data. The straw cover inversion results for study area 2 were similar to those for study area 1, with the four periods... The RMSE values ​​were 0.70, 0.75, 0.70, and 0.75, respectively. v The values ​​were 0.051, 0.044, 0.054, and 0.045, respectively, with the autumn data showing better accuracy than the spring data. Overall, the inversion accuracy obtained from the autumn data was significantly higher than that from the spring data, and the inversion accuracy from data of the same period across different study regions was similar. This result is consistent with the findings of the correlation analysis.

[0103] The evaluation index results of two study areas were analyzed in this case, and the local straw coverage estimation model constructed based on the RF algorithm optimized by GA-HPSO algorithm showed good results, indicating that processing and analyzing spectral reflectance information through machine learning algorithm can effectively extract features and improve the accuracy and estimation ability of straw coverage detection. The results were influenced by data at different times, and the inversion effect of autumn data was better than that of spring data, while different study areas had little effect on the results of straw coverage inversion model.

[0104] For step 4, this case used 8 sets of data collected, including 2 study areas and 4 periods. In order to test the sample experience distribution, the obtained samples were analyzed. The initial distribution of the sample was judged by looking at the sample curve in the histogram, and if the sample was near the straight line in the QQ(Quantile-Quantile) graph, it was proved that the sample basically met the normal distribution. Therefore, it is assumed that the global straw distribution is formed by a mixture of multiple Gaussian distributions, indicating that its probability model is Gaussian mixture model. When the global straw coverage probability model is assumed to be Gaussian mixture model, the expectation maximization algorithm is used to solve the parameters in the Gaussian mixture model. The total data sample is fitted into a mixed distribution as the global straw distribution probability model, and finally a certain number of samples are taken in the Gaussian mixture model to obtain the solution of the global coverage.

[0105] In order to evaluate the effectiveness of the Monte Carlo method proposed in this chapter, two evaluation indexes of accuracy and error are used. Since the distribution function of Gaussian mixture model cannot be solved, the direct sampling method cannot be used for random sampling, and the selection algorithm is used to randomly sample the Gaussian mixture model. The core idea of Monte Carlo method is random sampling, and the more sample values obtained, the higher the accuracy. Therefore, the number of iterations is set according to the total area and sampling area of the study area, and finally the total value of all samples that meet the probability distribution model is obtained as the solution of the global straw coverage. Based on the local estimation results, the global estimation is carried out by using Monte Carlo method, and the results are shown in Table 7 based on the global estimation results of Monte Carlo.

[0106] By comparing the results, it can be seen that the estimation results of Monte Carlo method reach the expected value, with an accuracy of 82.20% to 87.03%, and an average error of 1.10t per hectare. Considering the diversification of factors affecting straw coverage, the error is within an acceptable range. From the table, it can be found that the accuracy of Monte Carlo method in autumn is better than that in spring, with an average accuracy of 86.55% and an average error of 0.98t. In summary, the Monte Carlo estimation method based on local estimation results shows good accuracy and stability, and is persuasive, indicating that it is feasible to use Monte Carlo method to estimate large-scale field corn straw coverage based on sample.

[0107] Table 7 Global estimation results based on Monte Carlo

[0108]

[0109] The present application is based on the multi-spectral images obtained by the unmanned aerial vehicle, and the full-coverage sampling area is determined by visual observation method. The multi-spectral images are collected, and the corn straw coverage data is measured. The spectral index is selected as the input variable of the model. The RF algorithm and the GA-HPSO algorithm are used to construct the local estimation model of the corn straw coverage. The Monte Carlo sampling is used to estimate the global straw coverage, which provides a reference for the quantitative analysis of the straw coverage of the protective tillage unmanned aerial vehicle remote sensing. and The average reaches 0.77 and 0.72, and the RMSE v is only 0.050. At the same time, the Monte Carlo sampling can be used to estimate the large-scale corn straw coverage, and the quantitative goal of the straw returning to the field detection can be achieved. The accuracy in autumn reaches 86.55%, and the error is 0.98t, which provides a scientific and efficient automatic means for the detection and promotion of the protective tillage.

[0110] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which belong to the protection scope of the present application. Therefore, the protection scope of the present application patent should be subject to the appended claims.

Claims

1. A method for estimating the amount of straw mulch in a field, characterized in that, Includes the following steps: Acquire multispectral remote sensing images of each sampling point within the field study area, and obtain the straw coverage at each sampling point; Extract different spectral bands from the multispectral remote sensing image of each sampling point, and obtain the index values ​​of different spectral bands corresponding to each sampling point based on the correspondence between different spectral bands and vegetation indices; determine the correlation between the index values ​​of different spectral bands corresponding to each sampling point and the straw coverage of the corresponding sampling point. The straw coverage of each sampling point is sorted from small to large, and sampling points with strong correlation are selected. The index values ​​of different spectral bands corresponding to the selected sampling points are used to train the random forest model to learn the correlation between the index values ​​of different spectral bands and straw coverage. The random forest model is then optimized using a genetic-chaotic particle swarm optimization algorithm to obtain a model for estimating straw coverage. The index values ​​of different spectral bands corresponding to each sampling point in the field study area are input into the model used to estimate the straw coverage amount, the straw coverage amount of each sampling point is obtained, and the average value of the straw coverage amount of each sampling point is taken as the field straw coverage amount.

2. The method for estimating field straw mulch as described in claim 1, characterized in that, The method for obtaining the exponent values ​​of different spectral bands corresponding to each sampling point includes the following steps: The multispectral remote sensing images of each sampling point were band registered using remote sensing image processing tools, and geometric correction preprocessing was performed. The preprocessed multispectral remote sensing image of each sampling point is imported into remote sensing image processing software to extract spectral reflectance data of different spectral bands of each sampling point; Based on the spectral reflectance data of different spectral bands at each sampling point, the vegetation index solution method corresponding to different spectral bands is used to obtain the index value of different spectral bands for each sampling point.

3. The method for estimating field straw mulch as described in claim 1, characterized in that, The specific steps for obtaining the model used to estimate straw cover amount include: The straw coverage at each sampling point was sorted from smallest to largest, and divided into a modeling set and a validation set with a ratio of 2:1, and samples were taken at equal intervals. Sampling points with strong correlation between the index values ​​of different spectral bands corresponding to each sampling point and the straw coverage of the corresponding sampling points are selected. Random forest models are trained using the index values ​​of different spectral bands corresponding to the selected sampling points to obtain models for estimating straw coverage. The genetic-chaotic particle swarm optimization algorithm is then used to optimize the models for estimating straw coverage. The predicted and measured straw coverage at the sampling points were used as the validation set, and the accuracy was evaluated using the coefficient of determination and root mean square error.

4. The method for estimating field straw mulch as described in claim 1, characterized in that, The method for acquiring the multispectral remote sensing image of each sampling point includes the following specific steps: The field study area was divided into multiple sampling units on an average basis. Six sampling points were randomly selected in each unit, and each straw-covered sampling point was set as a square area of ​​0.5m × 0.5m. A drone equipped with a multispectral camera was used to perform vertical shooting at low altitude in manual mode to obtain multispectral remote sensing images of each sampling point.

5. The method for estimating field straw mulch as described in claim 1, characterized in that, The average straw coverage at each sampling point was estimated using a Monte Carlo model.

6. A device for estimating the amount of straw mulch in a field, characterized in that, include: The data acquisition module is used to acquire multispectral remote sensing images of each sampling point within the field research area and to obtain the straw coverage at each sampling point; The correlation analysis module is used to extract different spectral bands from the multispectral remote sensing image of each sampling point, and obtain the index values ​​of different spectral bands corresponding to each sampling point based on the correspondence between different spectral bands and vegetation indices; and determine the correlation between the index values ​​of different spectral bands corresponding to each sampling point and the straw coverage of the corresponding sampling point. The model building module is used to sort the straw coverage of each sampling point from small to large and select sampling points with strong correlation. The exponent values ​​of different spectral bands corresponding to the selected sampling points are used to train the random forest model to learn the correlation between the exponent values ​​of different spectral bands and straw coverage. The random forest model is then optimized using a genetic-chaotic particle swarm optimization algorithm to obtain a model for estimating straw coverage. The straw cover estimation module is used to input the index values ​​of different spectral bands corresponding to each sampling point in the field research area into the model for estimating straw cover, obtain the straw cover of each sampling point, and take the average straw cover of each sampling point as the field straw cover.

7. An electronic device, comprising: Memory and processor; The memory stores a computer program, characterized in that when the processor executes the computer program, it implements a method for estimating the amount of straw cover in the field according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements a method for estimating field straw coverage as described in any one of claims 1 to 5.

Images