CN116818688A

CN116818688A - Method and equipment for estimating soil salinity by unmanned aerial vehicle multispectral image

Info

Publication number: CN116818688A
Application number: CN202310774098.5A
Authority: CN
Inventors: 赵文举; 马芳芳
Original assignee: Lanzhou University of Technology
Current assignee: Lanzhou University of Technology
Priority date: 2023-06-28
Filing date: 2023-06-28
Publication date: 2023-09-29

Abstract

The application discloses a method and equipment for estimating soil salinity by using an unmanned aerial vehicle multispectral image, wherein the equipment used by the method comprises an unmanned aerial vehicle, a multispectral camera assembled on the unmanned aerial vehicle and sampling equipment for soil sampling; the sampling equipment comprises a sampling machine body, a cantilever mechanism is assembled on the sampling machine body, a rotating sheet is connected at the end part of the cantilever mechanism in a pin joint mode, a worm wheel box is welded on the rotating sheet, a worm wheel and a motor connected with the worm wheel are arranged in the worm wheel box, a worm is meshed in the worm wheel box, and the tail end of the worm is connected with a sampling drill rod. By utilizing the spectral information and the texture characteristics of the multispectral image, a learning model is built through double screening of sensitive variables, and compared with an inversion model built by a single spectral index, the combination of the texture characteristics and the spectral information improves the inversion model precision and inversion effect to a certain extent, and solves the problem of insufficient inversion precision of soil salinity covered by vegetation in a farming area.

Description

Method and equipment for estimating soil salinity by unmanned aerial vehicle multispectral image

Technical Field

The application relates to the technical field of soil parameter inversion, in particular to a method and equipment for estimating soil salinity by using a multispectral image of an unmanned aerial vehicle.

Background

The method can quickly and accurately evaluate the salt content of the soil at different depths under the crop coverage of the farming area, and has important significance for fully knowing the salt condition of the soil in the growth process of the crops, improving the salinization of the soil of the farming area and improving the yield of the crops. The unmanned plane has the advantages of simple operation, short duration, high flux and the like, is widely applied to soil parameter or crop growth monitoring, has the advantages that an expert extracts texture features and texture indexes of images for inversion research of vegetation ground biomass, and results show that the inversion accuracy of a model is remarkably improved after the texture information is added.

The existing unmanned aerial vehicle is used for soil salinity inversion research, the spectral indexes are basically used as input, the information of the unmanned aerial vehicle spectral images is ignored, model optimization still needs to be performed again when inversion is performed on different crops and covered soil in the growing period, the steps are repeated, the workload is high, and the problems that the precision of inversion models of the crop covered soil in the farming area is low, the work is heavy and the like are solved.

Disclosure of Invention

This section is intended to outline some aspects of embodiments of the application and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section as well as in the description of the application and in the title of the application, which may not be used to limit the scope of the application.

In view of the problems that in the prior art, inversion accuracy of salt content of soil with different depths is low due to the fact that soil in a farming area is blocked by crops, model optimization still needs to be conducted again when inversion is conducted on the soil covered by different crops and growth periods, steps are repeated, workload is high, and the problems that inversion model accuracy of the soil covered by crops in the farming area is low, work is heavy and the like are solved.

In order to solve the technical problems, the application provides the following technical scheme: the method for estimating soil salinity by using the multispectral image of the unmanned aerial vehicle specifically comprises the following steps:

step 1: carrying a multispectral camera by using an unmanned aerial vehicle to obtain multispectral images of a crop cultivation area; meanwhile, a sampling device is used for collecting a soil sample, and a conductivity meter is used for measuring the conductivity of the soil sample;

step 2: preprocessing such as correcting, splicing and the like is carried out on the multispectral image of the unmanned aerial vehicle;

step 3: extracting the reflectivity of each wave band of the multispectral image, and constructing a spectrum index;

step 4: calculating the vegetation coverage rate of the corresponding crops, memorizing the types of the crops, the spectral reflectivity and the vegetation coverage rate, and establishing a crop growth information base;

step 5: extracting texture features of the images of each wave band, and constructing texture indexes by using a spectrum index calculation formula with high correlation degree;

step 6: a support vector machine characteristic recursion elimination algorithm and gray correlation analysis are utilized to doubly screen sensitive variables;

step 7: constructing soil salinity inversion models with different depths by using 4 machine learning algorithms, namely an extreme learning machine, a support vector machine, a reverse neural network and a random forest;

step 8: performing performance estimation on each model modeling set and each verification set by adopting 3 evaluation indexes of a decision coefficient, a root mean square error and an average absolute error, and determining optimal estimation models of soil salinity of different depths of vegetation coverage;

step 9: and memorizing the determined optimal soil salinity estimation model in a crop information base.

As a preferable scheme of the method for estimating soil salinity by the multispectral image of the unmanned aerial vehicle, the application comprises the following steps: collecting blue light (B), green light (G), red light (R), near Infrared (NIR) and red edge (edge) in the step 1 for 5 spectrum bands; the soil sample is weighed, dried, ground, sieved, stirred with water and left to stand before the conductivity of the soil sample is measured by using a conductivity meter.

As a preferable scheme of the method for estimating soil salinity by the multispectral image of the unmanned aerial vehicle, the application comprises the following steps: in step 3, the reflectivities of blue light (B), green light (G), red light (R), near Infrared (NIR) and red edge (ridge) are respectively extracted, a spectral index is constructed, 16 salt indexes and vegetation indexes are included, the red edge band is sensitive to vegetation growth response, the red edge band is used for replacing the red light band, and each spectral index is improved.

As a preferable scheme of the method for estimating soil salinity by the multispectral image of the unmanned aerial vehicle, the application comprises the following steps: the growth information base in the step 4 mainly comprises a searching module, a memory module, an analysis module and a management module, can collect and count the growth information such as crop type, spectral reflectivity, vegetation coverage rate and the like, can memorize the optimal inversion model of the existing crops, and can quickly select the proper inversion model according to crop varieties and growth periods in the later period.

As a preferable scheme of the method for estimating soil salinity by the multispectral image of the unmanned aerial vehicle, the application comprises the following steps: in step 5, gray level co-occurrence matrix is a more common method for extracting texture features, and 8 texture features in the spectral bands of the multispectral image are extracted by using gray level co-occurrence matrix method respectively: the method comprises the steps of sorting all salinity indexes, vegetation indexes by gray correlation analysis, selecting two of 40 texture features in 5 wave bands based on a spectrum index idea, and constructing a texture index by a spectrum index calculation formula with highest correlation degree.

As a preferable scheme of the method for estimating soil salinity by the multispectral image of the unmanned aerial vehicle, the application comprises the following steps: in the step 9, the type of crops, the spectral reflectivity and the vegetation coverage rate are compared through a crop information base, and when the salinity inversion is performed in the next year or in the next round of growth of the crops, the recorded optimal inversion model can be directly selected for the salinity inversion in the same growth period of the same kind of crops, so that the purpose of rapidly and efficiently estimating the salinity of the vegetation coverage soil in the farming area is achieved.

The method for estimating soil salinity by using the multispectral image of the unmanned aerial vehicle has the beneficial effects that:

according to the method for estimating soil salinity by using the multispectral image of the unmanned aerial vehicle, provided by the application, the spectral information and the texture characteristics of the multispectral image are utilized, the machine learning model based on the extreme learning machine, the support vector machine, the reverse neural network and the random forest is constructed by doubly screening the sensitive variables, and compared with the inversion model established by a single spectral index, the inversion model precision and inversion effect are improved to a certain extent by combining the texture characteristics and the spectral information, and the problem of insufficient inversion precision of vegetation coverage soil salinity in a farming area is solved.

The crop information database which is specially used for inversion of crop coverage soil in a farming area and contains information such as crop varieties, spectral reflectivity, coverage rate and the like is provided, the information database mainly comprises a search module, a memory module, a management module and an analysis module, crop growth information can be collected and counted, a proper inversion model can be quickly selected according to the crop varieties and the growth period, other soil salt content inversion tasks can be trained, and the workload of actual measurement sampling and the difficulty of model selection are greatly reduced.

The application also provides another technical scheme, which mainly provides corresponding shooting for a method for estimating soil salinity by using the multispectral image of the unmanned aerial vehicle and spectrum shooting and sampling for soil by using sampling equipment.

In order to solve the technical problems, the application provides the following technical scheme: the equipment for estimating soil salinity by using the multispectral image of the unmanned aerial vehicle specifically comprises the unmanned aerial vehicle, a multispectral camera assembled on the unmanned aerial vehicle and sampling equipment for sampling soil,

the multispectral camera is provided with the cradle head, so that the multispectral camera is convenient to connect with the unmanned aerial vehicle;

the sampling equipment comprises a sampler main body, a cantilever mechanism is assembled on the sampler main body, a rotating sheet is connected at the end part of the cantilever mechanism in a pin joint mode, a worm wheel box is welded on the rotating sheet, a worm wheel and a motor connected with the worm wheel are arranged in the worm wheel box, a worm is meshed in the worm wheel box, the tail end of the worm is connected with a sampling drill rod, a second hydraulic rod is connected at the lower end of the rotating sheet in a pin joint mode, and the second hydraulic rod is arranged below the cantilever mechanism.

As a preferable scheme of the method for estimating soil salinity by the multispectral image of the unmanned aerial vehicle, the application comprises the following steps: the cantilever mechanism comprises a big arm and a small arm, wherein the small arm is slidably connected in the big arm, the end part of the small arm is welded with a connecting frame, a rotating piece is installed at the end part of the connecting frame through a rotating shaft, a second hydraulic rod is assembled at the lower end of the connecting frame, a connecting angle steel is assembled on the lower surface of the rear end of the big arm, a supporting rod is pinned on a sampler main body, the big arm is connected with the supporting rod through the connecting angle steel, a telescopic rod is further arranged on the sampler main body and is pinned with the connecting angle steel, a handle is further arranged at the rear end of the upper surface of the supporting rod, a first hydraulic rod is assembled on the upper surface of the big arm, and the output end of the first hydraulic rod is pinned with the top of the connecting frame.

As a preferable scheme of the method for estimating soil salinity by the multispectral image of the unmanned aerial vehicle, the application comprises the following steps: the sampling drill rod is of a hollow structure, a bulldozing piece is arranged in the sampling drill rod, the bulldozing piece comprises a disc-shaped bulldozing piece with the diameter equal to that of the inner wall of the sampling drill rod and a push rod connected to the bulldozing piece, a push button is integrally connected to the side face of the upper end of the push rod, and a through groove for pushing and twisting is formed in the side face of the sampling drill rod.

The unmanned aerial vehicle multispectral image provided by the application has the beneficial effects on soil salinity estimation equipment:

the sampling device can sample soil in a rape and aversion area, the whole sampling device is of a trolley structure, an operator can conveniently carry the device to move, a cantilever mechanism arranged on the sampling device can stretch and prolong, so that multipoint sampling can be carried out by taking the device as a center under the condition of not moving the device, the angle of a sampling drill rod for sampling can be adjusted through a rotating plate and a second hydraulic rod which are assembled on the device, sampling at different angles is realized, the sampling drill rod uniformly descends through rotation of a worm gear so as to puncture the soil for sampling, sampling at different depths is realized, and a bulldozing piece arranged in the sampling drill rod can completely push out the sample in the sampling drill rod, so that the integrity of the sample is ensured.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art. Wherein:

fig. 1 is a diagram of the structure of a multi-spectral camera-loaded drone.

Fig. 2 is a diagram showing a connection structure of a multispectral camera and a cradle head.

Fig. 3 is a block diagram of a sampling device.

Fig. 4 is a structural view of the cantilever mechanism.

Fig. 5 is an enlarged view of the structure at a in fig. 3.

FIG. 6 is a cross-sectional view of a sampling drill.

FIG. 7 is a flow chart of a soil salinity inversion method.

Fig. 8 is a flow chart of the operation of the crop information library.

The correspondence between the reference numerals and the component names in the drawings is as follows: 100. unmanned plane; 101. a cradle head; 102. a multispectral camera; 200. a sampler body; 200a, a handle; 200b, stay bars; 200c, a telescopic rod; 201. a cantilever mechanism; 201a, large arm; 201a-1, forearm; 201a-2, a first hydraulic lever; 201b, connecting angle steel; 202. a connecting frame; 202a, a second hydraulic rod; 202b, rotating the sheet; 202b-1, a worm gear box; 202b-1a, a worm; 202b-1b, sampling drill rod; 202b-1b-1, through slots; 202b-1b-2, bulldozer; 202b-1b-2a, push button.

Detailed Description

In order that the above-recited objects, features and advantages of the present application will become more readily apparent, a more particular description of the application will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application, but the present application may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present application is not limited to the specific embodiments disclosed below.

Further, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic can be included in at least one implementation of the application. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.

Example 1

Referring to fig. 7 and 8, in embodiment 1 of the present application, a method for estimating soil salinity by using multispectral images of an unmanned aerial vehicle is provided, and by using memory storage, an optimal inversion model can be rapidly selected according to crop varieties and growth periods, so as to realize accurate and efficient inversion of soil salinity in a farming area, and specifically comprises the following steps:

Referring to fig. 1, in step 1: and acquiring multispectral images of a crop cultivation area by using an unmanned aerial vehicle, and collecting 5 spectral bands of blue light (B), green light (G), red light (R), near Infrared (NIR) and red edge (edge). Simultaneously, sampling soil with different depths under the coverage of crops, uniformly distributing sampling points in a crop planting area by adopting a five-point method, respectively collecting 0-20 cm, 20-40 cm and 40-60cm of soil, wherein 0-60cm is a distribution area of a main root system of a general crop.

In the step 1, the collected soil sample is weighed, dried, ground and sieved, distilled water with the mass of 5 times of the soil sample is added for full stirring and standing, the conductivity of the soil sample is measured by using a conductivity meter, and the conductivity is converted into the salt content of the soil, wherein the salt content of the soil is calculated in percent.

Step 3: and respectively extracting the reflectivities of blue light (B), green light (G), red light (R), near Infrared (NIR) and red edge (R-edge), constructing a spectrum index, wherein the spectrum index comprises 16 salt indexes and vegetation indexes, the red edge wave band is sensitive to vegetation growth response, and the red edge wave band is used for replacing the red light wave band, so that each spectrum index is improved.

It should be noted that: in the above embodiment, the red-side band is used to replace the red-side band, and the calculation formulas of the improved salinity index and the vegetation index of the red-side band are as follows:

normalized salt index ndsi= (ridge-NIR)/(ridge+nir)

Ratio vegetation index rvi=nir/edge

Differential vegetation index dvi=nir-edge

Green leaf index gli= [ (G-ridge) + (G-B) ]/2g+ridge+b

Enhanced vegetation index evi=2.5 (NIR-edge)/(nir+6edge-7.5b+1)

Leaf area index lai=3.618 [2.5 (NIR-edge)/(nir+6edge-7.5b+1) ] -0.118

Soil improvement vegetation index msavi=2ridge+1- [ (2nir+1) ² -8(NIR-Redge)] ^0.5 /2

Nonlinear index nli= (NIR ² -Redge)/(NIR ² +Redge)

Normalized salt index ndvi= (NIR-ridge)/(nir+ridge)

Salt index si= (b+ridge) ^0.5

Salt index si1= (g×ridge) ^0.5

Salt index si2= (G) ² +Redge ² +NIR ² ) ^0.5

Salt index SI-t=100 (ridge/NIR)

Intensity index int1= (g+ridge)/2

Intensity index Int2= (G+Reg+NIR)/2

Luminance index bi= (ridge+nir) ^0.5

Wherein: B. g, R, NIR, R-edge are the spectral reflectivities of the blue, green, red, near infrared and red bands, respectively.

Step 4: the varieties, the ground object spectral reflectivities and the vegetation coverage of inversion crops are memorized, a crop growth information base is established, the types and the growth periods of the crops can be conveniently and quickly identified in the next year or in the next round of growth, data are provided for quickly determining the most suitable soil salinity inversion model, specifically, the vegetation coverage is greatly different along with the different growth periods of the crops and the different varieties of the crops, the ground object supervision classification can be realized in a Supervised Classification window of ENVI5.3, samples of crops, soil, weeds and the like in a farming area are firstly established, the separability among the samples is calculated, when the separability is larger than 1.9, the two types of samples can be directly distinguished, then a neural network supervision classification (Neural Network Classification) method is selected in a Supervised Classification window, the ground object classification is realized, and the classified crop area occupies the land area and is the vegetation coverage. And (3) establishing a crop growth information base, and memorizing the crop type, the crop spectral reflectivity and the vegetation coverage rate into the crop information base so as to monitor and invert the salinity of the covered soil of the same crop and the same fertility period next time.

The crop information base is mainly characterized by mainly comprising a searching module, a memory module, an analysis module and a management module, wherein the crop information base can collect and count the growth information such as crop type, spectral reflectivity, vegetation coverage rate and the like, can memorize the optimal inversion model of the existing crops, and can quickly select the proper inversion model according to crop varieties and growth periods in the later period.

Step 5: the gray level co-occurrence matrix is a more common method for extracting texture features, and 40 texture features of 5 spectral bands of the multispectral image are respectively extracted by using a gray level co-occurrence matrix method: the method is specifically implemented in a CO-Occurrence Measuers window of ENVI5.3 or MATLAB R2017B, and respectively extracts texture features of blue light (B), green light (G), red light (R), near Infrared (NIR) and red edge (edge) in total of 5 spectral bands, wherein the texture features comprise 8 texture features of mean, variance, cooperativity, contrast, dissimilarity, entropy, second moment and relativity. The calculation formula is as follows:

mean value of

Variance of

Cooperativity of

Contrast ratio

Dissimilarity and dissimilarity

Entropy of

Second moment of

Correlation of

Wherein: i and j represent gray values at the ith row and the jth column of the spectral image, respectively; p (P) _i,j Representing the probability of occurrence of the gray value corresponding to the ith row and the jth column in the matrix; n is the gray level of the image; mu (mu) _i Sum mu _j Sigma is the mean value of gray level co-occurrence matrix _i Sum sigma _j Is the standard deviation.

Step 6: and (5) performing double screening on the sensitive variables by using a support vector machine characteristic recursion elimination algorithm and gray correlation analysis. According to the similarity or dissimilarity degree of the development trends among the elements, the gray correlation degree among the variables is calculated, the variables above a certain threshold value are taken as input, the SVM-RFE mainly utilizes an SVM algorithm to evaluate the importance degree of each characteristic variable, and the variables with low importance are removed one by one according to backward iteration, compared with a correlation screening method, the algorithm can effectively avoid the condition that part of the variables are filtered before entering a model, so that the double screening method provided by the application can effectively screen out the variables sensitive to the response of the measured soil salt content; specifically, variable screening is realized in MATLAB R2017b, 8 SVM-RFE reserved variables are set, variable screening is carried out on salinity index, vegetation index, texture feature and texture index, gray relevance ranking is carried out on the 8 screened variables, and 5 sensitive variables with the forefront ranking are taken as model input.

Step 7: the dataset was set at 7:3, dividing the soil salinity inversion model into a modeling set and a prediction set, and constructing soil salinity inversion models with different depths under crop coverage by using an extreme learning machine, a support vector machine, a reverse neural network and a random forest in total of 4 machine learning algorithms.

Step 8: and 3 evaluation indexes of the decision coefficient, the root mean square error and the average absolute error are adopted to carry out performance estimation on each model modeling set and each verification set, the optimal estimation model of the soil salinity of different depths of vegetation coverage is determined, a model algorithm is stored in a crop growth information base together, the type and the growth period of crops can be rapidly identified by soil salinity inversion of the next year or the next rotation, and the salinity inversion is directly carried out according to the stored optimal soil salinity inversion model.

Specifically, the calculation formulas of the determination coefficient, the root mean square error and the average absolute error are as follows:

wherein: y is _i For measured soil salt values,%;to predict soil salinity values,%;Soil salinity mean,%; n is the number of samples.

R ² The closer to 1 the model accuracy is, the closer to 0 the RMSE is, the higher the model accuracy is, the closer to 0 the MAE value is, and the higher the model prediction accuracy is.

In the step 9, the crop type, the spectral reflectivity and the vegetation coverage rate are compared through the crop information base, and the crop type, the spectral reflectivity and the vegetation coverage rate can be directly compared through the crop information base in the next year or in the next crop rotation growth process, and the recorded optimal inversion model can be directly selected for carrying out the salinity inversion in the same growth period of the same crop, so that the purpose of rapidly and efficiently estimating the salinity of the vegetation coverage soil of a farming area is achieved.

In this embodiment: according to the method for estimating the soil salinity by the multispectral image of the unmanned aerial vehicle, the spectral information and the texture characteristics of the multispectral image are utilized, the machine learning model based on the extreme learning machine, the support vector machine, the reverse neural network and the random forest is constructed through double screening of sensitive variables, compared with the inversion model established by a single spectral index, the combination of the texture characteristics and the spectral information improves the inversion model precision and the inversion effect to a certain extent, the problem of insufficient inversion precision of the soil salinity covered by vegetation cover soil in a farming area is solved, a crop information database which is specially used for inversion of the crop cover soil in the farming area and contains information such as crop varieties, spectral reflectivity, coverage rate and the like is provided, the information database mainly comprises a search module, a memory module, a management module and an analysis module, the crop growth information can be collected and counted, a proper inversion model can be selected quickly according to the crop varieties and the growth period, other soil salinity inversion tasks can be trained, and the actual measurement sampling workload and the model selection difficulty are greatly reduced.

Example 2

Referring to fig. 1-6, in the embodiment 2 of the present application, by setting corresponding devices to collect remote sensing images and soil samples, specifically, an apparatus for estimating soil salinity by using multispectral images of an unmanned aerial vehicle includes an unmanned aerial vehicle 100 for performing aerial view large-area image collection, a multispectral camera 102 mounted on the unmanned aerial vehicle 100 for shooting, and a sampling apparatus for soil sampling, where a pan-tilt 101 is mounted on the multispectral camera 102 to facilitate connection of the multispectral camera 102 with the unmanned aerial vehicle 100, so that the multispectral camera 102 is sent into the air for shooting by using the unmanned aerial vehicle 100, and the pan-tilt 101 is used in the prior art for adjusting the angle of the multispectral camera 102;

in fig. 3 to 6, the sampling apparatus includes a cart-shaped sampling machine body 200, a cantilever mechanism 201 capable of extending a sampling range is mounted on the sampling machine body 200, a rotation piece 202b for adjusting an angle is pin-connected to an end portion of the cantilever mechanism 201, a worm wheel box 202b-1 is welded to the rotation piece 202b, a worm wheel and a motor connected to the worm wheel are provided in the worm wheel box 202b-1, a worm 202b-1a is engaged in the worm wheel box 202b-1, a sampling drill rod 202b-1b is connected to an end portion of the worm 202b-1a, a second hydraulic rod 202a is pin-connected to a lower end portion of the rotation piece 202b, and the second hydraulic rod 202a is mounted under the cantilever mechanism 201, the worm 202b-1 is driven to rise and fall by the worm wheel in the worm wheel box 202b-1, the sampling drill rod 202a integrally connected to a bottom portion is penetrated into the ground to sample, and the sampling drill rod 202b is driven to rotate by expansion and contraction to an angle at the rotation piece 202b at an end portion of the cantilever mechanism 201, thereby adjusting a sampling angle of the sampling drill rod 202 b-1.

In fig. 4 and 5, the cantilever mechanism 201 includes a fixed large arm 201a and a telescopic small arm 201a-1, wherein the small arm 201a-1 is slidably connected in the large arm 201a, a connecting frame 202 is welded at the end of the small arm 201a-1, and a rotating plate 202b is mounted at the end of the connecting frame 202 by a rotating shaft, a second hydraulic rod 202a is assembled at the lower end of the connecting frame 202, the connecting frame 202 is used for loading the rotating plate 202b and adjusting the second hydraulic rod 202a of the rotating plate 202b, a connecting angle steel 201b is assembled at the lower surface of the rear end of the large arm 201a, a supporting rod 200b is pinned on the sampler body 200, the large arm 201a is connected with the supporting rod 200b by the connecting angle steel 201b, a telescopic rod 200c is pinned with the connecting angle steel 201b, the large arm 201a is controlled to make lever motion on the supporting rod 200b by pushing and pulling the connecting angle steel 200c, thereby realizing the lifting and lowering of the cantilever mechanism 201, the lifting arm mechanism can be assembled at the upper end of the sampling rod 201a by the first hydraulic rod 201a, the lifting range of the large arm 201a can be further connected with the sampling rod 201a by the first hydraulic rod 2, and the lifting arm 201a can be further connected at the upper end of the sampling rod 201a 2 by the upper end of the connecting rod 201 a.

In fig. 6, a sampling drill 202b-1b is designed to be hollow for storing a soil sample, a bulldozing piece 202b-1b-2 is arranged in the sampling drill 202b-1b, the bulldozing piece 202b-1b-2 comprises a disc-shaped bulldozing piece with the same diameter as the inner wall of the sampling drill 202b-1b and a push rod connected to the bulldozing piece, wherein the push button 202b-1b-2a is integrally connected to the upper side surface of the push rod, a through groove 202b-1b-1 for the push button 202b-1b-2a to move is formed in the side surface of the sampling drill 202b-1b, after the sampling drill 202b-1b pierces the soil, the soil is sampled into the hollow of the sampling drill 202b-1b, the soil sample is stored in the sampling drill 202b-1b after the sampling drill 202b-1b is pulled out, the sample cannot slide out of the sampling drill 202b-1b due to friction and humidity problems of the sample, at this time, the push button 202b-1b-2a is pushed down by the external force to the push button 202b-1b-1 a moves downwards in the through groove 202b-1b-1 a, and the push rod 202b-1b-2a moves synchronously to push rod from the disc-shaped soil sample 202b-1 b.

In this embodiment: the multispectral camera 102 is carried on the cradle head 101 on the unmanned aerial vehicle 100 to acquire multispectral images of a crop cultivation area, the sampler main body 200 is pushed to a preset area by the aid of the handle 200a in a synchronous mode, the telescopic rod 200c is controlled to lift the large arm 201a on the connecting angle steel 201b, the large arm 201a is installed on the supporting rod 200b, so that the front end of the large arm 201a is lifted up to span an obstacle, at the moment, the second hydraulic rod 202a pulls the rotating piece 202b to rotate to adjust the sampling drill rod 202b-1b to a vertical state, then the first hydraulic rod 201a-2 pushes the small arm 201a-1 to extend out of the large arm 201a, the whole adjustment of the cantilever mechanism 201 is completed, the radiation range of sampling is prolonged, the worm wheel in the worm wheel box 202b-1 is used to drive the worm 202b-1 to lift, the sampling drill rod 202b-1b integrally connected with the bottom is penetrated into the ground to sample when the worm 202b-1 descends, a soil sample is stored in the sampling drill rod 202b-1b due to friction and humidity problems that the sample 202b-1b is pushed out of the sample 202b-1b is not pushed out of the sample 202b-1b-1 because of the sample is pushed out of the sample 202b-1 b-1.

It is important to note that the construction and arrangement of the application as shown in the various exemplary embodiments is illustrative only. Although only a few embodiments have been described in detail in this disclosure, those skilled in the art who review this disclosure will readily appreciate that many modifications are possible (e.g., variations in sizes, dimensions, structures, shapes and proportions of the various elements, values of parameters (e.g., temperature, pressure, etc.), mounting arrangements, use of materials, colors, orientations, etc.) without materially departing from the novel teachings and advantages of the subject matter described in this application. For example, elements shown as integrally formed may be constructed of multiple parts or elements, the position of elements may be reversed or otherwise varied, and the nature or number of discrete elements or positions may be altered or varied. Accordingly, all such modifications are intended to be included within the scope of present application. The order or sequence of any process or method steps may be varied or re-sequenced according to alternative embodiments. In the claims, any means-plus-function clause is intended to cover the structures described herein as performing the function and not only structural equivalents but also equivalent structures. Other substitutions, modifications, changes and omissions may be made in the design, operating conditions and arrangement of the exemplary embodiments without departing from the scope of the present applications. Therefore, the application is not limited to the specific embodiments, but extends to various modifications that nevertheless fall within the scope of the appended claims.

Furthermore, in order to provide a concise description of the exemplary embodiments, all features of an actual implementation may not be described (i.e., those not associated with the best mode presently contemplated for carrying out the application, or those not associated with practicing the application).

It should be appreciated that in the development of any such actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort might be complex and time consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill having the benefit of this disclosure.

It should be noted that the above embodiments are only for illustrating the technical solution of the present application and not for limiting the same, and although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present application may be modified or substituted without departing from the spirit and scope of the technical solution of the present application, which is intended to be covered in the scope of the claims of the present application.

Claims

1. A method for estimating soil salinity using UAV multispectral imagery, characterized by the following steps:

Step 1: Use a drone equipped with a multispectral camera to acquire multispectral images of the crop cultivation area; at the same time, use sampling equipment to collect soil samples and use a conductivity meter to measure the conductivity of the soil samples;

Step 2: Perform preprocessing such as correction and stitching on the UAV multispectral images;

Step 3: Extract the reflectance of each band of the multispectral image and construct the spectral index;

Step 4: Calculate the corresponding crop vegetation coverage, and memorize the crop type, spectral reflectance, and vegetation coverage to establish a crop growth information database;

Step 5: Extract texture features from images of each band and construct texture indices using spectral index calculation formulas with high correlation.

Step 6: Use the support vector machine feature recursive elimination algorithm and grey relational analysis to screen sensitive variables;

Step 7: Construct soil salinity inversion models at different depths using four machine learning algorithms: Extreme Learning Machine, Support Vector Machine, Reverse Neural Network, and Random Forest;

Step 8: Use three evaluation indicators—coefficient of determination, root mean square error, and mean absolute error—to evaluate the performance of each model on the modeling set and validation set, and determine the optimal estimation model for soil salinity at different depths of vegetation cover.

Step 9: Memorize the determined optimal soil salinity estimation model in the crop information database.

2. The method for estimating soil salinity using UAV multispectral imagery as described in claim 1, characterized in that: in step 1, five spectral bands are acquired: blue light (B), green light (G), red light (R), near-infrared (NIR), and red edge (Redge); before the soil sample is used to determine the conductivity of the soil sample using a conductivity meter, it needs to be weighed, dried, ground, sieved, mixed with water, and left to stand.

3. The method for estimating soil salinity using UAV multispectral imagery as described in claim 1, characterized in that: in step 3, the reflectance of blue (B), green (G), red (R), near-infrared (NIR), and red-edge bands are extracted respectively to construct spectral indices, including 16 indices in total, including salinity index and vegetation index. The red-edge band is sensitive to vegetation growth, and the red-edge band is used to replace the red light band to improve each spectral index.

4. The method for estimating soil salinity using UAV multispectral imagery as described in claim 1, characterized in that: in step 4, the growth information database mainly includes a search module, a memory module, an analysis module, and a management module, which can collect and statistically analyze growth information such as crop type, spectral reflectance, and vegetation coverage, and can also memorize the most suitable inversion model for existing crops, so that the most suitable inversion model can be quickly selected based on crop variety and growth stage in the later stage.

5. The method for estimating soil salinity using UAV multispectral imagery as described in claim 1, characterized in that: in step 5, gray-level co-occurrence matrix is a commonly used method for extracting texture features. The gray-level co-occurrence matrix method is used to extract eight texture features of the spectral bands of the multispectral image: mean, variance, coherence, contrast, dissimilarity, entropy, second moment, and correlation. Gray relational analysis is used to rank the salinity index and vegetation index. Based on the idea of spectral index, two of the 40 texture features in 5 bands are randomly selected, and the texture index is constructed using the formula for calculating the spectral index with the highest correlation.

6. The method for estimating soil salinity using UAV multispectral imagery as described in claim 1, characterized in that: in step 9, by comparing crop type, spectral reflectance, and vegetation cover through a crop information database, when inverting salinity in the following year or during the next crop growth cycle, the best recorded inversion model can be directly selected for salinity inversion for the same crop at the same growth stage, thereby achieving the goal of rapidly and efficiently estimating soil salinity in agricultural areas with vegetation cover.

7. A device for estimating soil salinity using UAV multispectral imagery, applied to the method for estimating soil salinity using UAV multispectral imagery as described in claim 1, for spectral imaging and sampling of soil, comprising a UAV (100), a multispectral camera (102) mounted on the UAV (100), and sampling equipment for soil sampling, characterized in that:

The multispectral camera (102) is equipped with a gimbal (101) to facilitate the connection between the multispectral camera (102) and the drone (100);

The sampling equipment includes a sampling machine body (200), a cantilever mechanism (201) is mounted on the sampling machine body (200), a rotating plate (202b) is pinned to the end of the cantilever mechanism (201), a worm gear box (202b-1) is welded on the rotating plate (202b), a worm gear and a motor connected to the worm gear are provided in the worm gear box (202b-1), a worm (202b-1a) is meshed in the worm gear box (202b-1), a sampling probe (202b-1b) is connected to the end of the worm (202b-1a), a second hydraulic rod (202a) is pinned to the lower end of the rotating plate (202b), and the second hydraulic rod (202a) is installed below the cantilever mechanism (201).

8. The device for estimating soil salinity using UAV multispectral imagery as described in claim 7, characterized in that: the cantilever mechanism (201) includes a large arm (201a) and a small arm (201a-1), wherein the small arm (201a-1) is slidably connected within the large arm (201a), a connecting frame (202) is welded to the end of the small arm (201a-1), and a rotating plate (202b) is mounted on the end of the connecting frame (202) via a rotating shaft; the second hydraulic rod (202a) is assembled at the lower end of the connecting frame (202), and a [missing information - likely a component or part] is assembled on the lower surface of the rear end of the large arm (201a). A connecting angle steel (201b) is used, and a support rod (200b) is pinned to the main body (200) of the sampler. The boom (201a) is connected to the support rod (200b) via the connecting angle steel (201b). A telescopic rod (200c) is also provided on the main body (200) of the sampler and pinned to the connecting angle steel (201b). A handle (200a) is also provided at the rear end of the upper surface of the support rod (200b). A first hydraulic rod (201a-2) is mounted on the upper surface of the boom (201a), and the output end of the first hydraulic rod (201a-2) is pinned to the top of the connecting frame (202).

9. The device for estimating soil salinity using UAV multispectral imagery as described in claim 7, characterized in that: the sampling probe (202b-1b) is a hollow structure, and a bulldozing component (202b-1b-2) is built into the sampling probe (202b-1b). The bulldozing component (202b-1b-2) includes a disc-shaped bulldozing blade with the same diameter as the inner wall of the sampling probe (202b-1b) and a push rod connected to the bulldozing blade. A pusher (202b-1b-2a) is integrally connected to the upper side of the push rod, and a through groove (202b-1b-1) is provided on the side of the sampling probe (202b-1b) for the pusher (202b-1b-2a) to move.