Saline-alkali cultivated land salt spot extraction method and system based on multispectral unmanned aerial vehicle

Abstract

The invention relates to the technical field of salt spot recognition and extraction, in particular to a salt spot extraction method and system for saline-alkali cultivated land based on a multispectral unmanned aerial vehicle, and the method comprises the steps: obtaining image data through the unmanned aerial vehicle, and carrying out the splicing processing; calculating vegetation indexes and elevation data in different periods; screening a vegetation index through a relevancy screening model, and fitting and calculating a crop growth index leaf area index; coupling a crop growth index leaf area index and elevation data through a yield fitting model to obtain yield fitting data, and converting the yield fitting data into a visual result; and automatically extracting a salt spot range in combination with a machine learning algorithm. The invention further discloses a system for executing the method. According to the salt spot extraction method and system for the saline-alkali cultivated land based on the multispectral unmanned aerial vehicle, the salt spot range is extracted through coupling of the vegetation index and the elevation data, the extraction precision is greatly improved, complex image processing and coupling are not needed, the model result is visualized, the practicability is high, and prerequisite conditions are provided for subsequent application and popularization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of salt spot identification and extraction, and particularly relates to a salt spot extraction method and system for saline-alkali farmland based on a multispectral unmanned aerial vehicle. BACKGROUND

[0002] The saline-alkali land refers to a type of land where plants grow with difficulty due to excessive accumulation of salt in the soil. The farmland developed from the saline-alkali land through improvement is referred to as saline-alkali farmland. Due to the harm of soil salinity, in the process of developing and cultivating the saline-alkali land, it is inevitable that crops cannot grow in the salt patch area with patchy enrichment of salt, forming the salt spot phenomenon. In the current utilization and cultivation of the saline-alkali land, regardless of fertilization, pesticide application or irrigation, the land in the salt spot area and the surrounding ordinary land are cultivated in the same way, and the fertilizer type, quantity, fertilization method and the like based on the surrounding normal land are applied to the salt spot area with high salt content and poor soil structure, which has little effect on the vegetation in the salt spot area. It is almost inevitable that there are few seedlings in the salt spot area, and even if there are some weak seedlings in the early stage, they cannot form yield in the later stage. In this way, the fertilizer, pesticide and irrigation water applied to the salt spot area are completely wasted, and the salt spot cannot be effectively improved. If special fertilizer types, improvers and biological fertilizer based on the salt spot are applied, the cost of the whole land cultivation will be greatly increased, which is difficult for farmers to bear. Therefore, there is an urgent need for a salt spot accurate identification and positioning technology, which can be combined with agricultural machinery to carry out precise fertilization and pesticide application, and significantly improve the improvement efficiency of the saline-alkali farmland.

[0003] In addition, in the evaluation of the improvement effect of the saline-alkali land, there are two common yield measurement methods. One is actual yield measurement, which harvests all crops in the land, obtains the actual yield, and then divides the yield by the area of the whole land to obtain the yield per mu. The other is sample point yield measurement, which selects a place with uniform growth as a sample point, measures the actual yield in a small area, and then divides the yield by the area of the sample point to obtain the yield per mu. If the measurement is accurate, multiple sample points can be taken, and the average yield per mu of the sample points is taken as the yield of the land. However, the existence of the salt spot greatly interferes with the measurement and accuracy of the actual yield of the saline-alkali farmland. In the actual yield measurement, the area of the salt spot is irregular, and the area of the salt spot cannot be deducted when calculating the area of the whole land, so the measured yield is seriously inconsistent with the actual growth of the crops, and has little practical significance. In the sample point yield measurement, it is basically impossible to select sample points scientifically, reasonably and representatively due to the existence of a large number of salt spots with different sizes and properties, and the accuracy of the sample point yield measurement cannot be discussed.

[0004] The unmanned aerial vehicle with a camera has gradually become a modern tool for agricultural field patrol due to its flexibility and high efficiency in aerial survey. The prior art often directly performs graphic classification on visible light images or multispectral data through machine learning to identify the salinity degree of saline cultivated land. However, since the visible light and multispectral data contain a large amount of information, direct classification and identification cannot eliminate redundant information, resulting in low recognition accuracy, complex algorithm, and limited coverage, which cannot provide accurate data support for subsequent agricultural operations and model promotion. SUMMARY

[0005] The purpose of the present application is to provide a saline cultivated land salt spot extraction method and system based on a multispectral unmanned aerial vehicle, which filters the multispectral and elevation information obtained by the unmanned aerial vehicle, optimizes the main information, and makes the application of machine learning algorithm classification or numerical classification method more accurate and efficient, thereby providing accurate digital basis for the "symptomatic treatment" of salt spots in saline cultivated land.

[0006] To achieve the above purpose, the present application provides a saline cultivated land salt spot extraction method based on a multispectral unmanned aerial vehicle, and the specific steps are as follows: Step S1: An unmanned aerial vehicle equipped with a multispectral camera and a real-time dynamic positioning module is used to collect image data of saline cultivated land crops at different stages; Step S2: The images collected at each stage are spliced to obtain a complete saline cultivated land image; Step S3: Perform band calculation on the complete saline cultivated land image data to obtain vegetation index and elevation data at different stages; Step S4: Establish a correlation screening model, determine the vegetation index most correlated with the crop growth index leaf area index according to the correlation degree through the correlation screening model, and calculate the crop growth index leaf area index through the vegetation index most correlated with the previous growth period of the crop mature period; Step S5: Establish a yield fitting model, couple the crop growth index leaf area index and the corresponding elevation data through the yield fitting model, obtain yield fitting data, and convert it into a visual result; Step S6: Use the visual result as the base map data for image classification, and automatically extract the salt spot range combined with a machine learning algorithm.

[0007] Preferably, in step S1, the multispectral camera obtains at least red light band images, green light band images, near-infrared band images, and red edge band images; and the unmanned aerial vehicle is provided with a real-time dynamic positioning module for obtaining the altitude elevation value of the crop or cultivated land surface; The salt-alkali farmland is photographed by the unmanned aerial vehicle before the crops are sowed and germinated and in the previous growth period of the mature period of the crops, an initial digital surface model of the bare land of the salt-alkali farmland is obtained, initial elevation values at the time of the bare land, elevation values at the time of the maximum plant height and vegetation indexes in the previous growth period of the mature period of the crops are obtained; When the digital surface model of the salt-alkali farmland is obtained, in order to improve the accuracy of the elevation values, the unmanned aerial vehicle adopts the way of inclined flight to collect the elevation values at the time of the maximum plant height and the time of the bare land.

[0008] Preferably, in step S3, the vegetation indexes include normalized difference vegetation index, green normalized difference vegetation index, soil-adjusted vegetation index, optimized soil-adjusted vegetation index, normalized difference red edge index, improved chlorophyll absorption index and green red vegetation index.

[0009] Preferably, the plant height data is calculated according to the difference between the elevation values at the time of the maximum plant height and the initial elevation values at the time of the bare land, and is fitted by monadic linear regression of the plant height measured by the fixed point sample, and the fitting calculation formula is as follows: ; Wherein, is the plant height data after fitting, and are the first plant height fitting coefficient and the second plant height fitting coefficient respectively, is the difference between the elevation values at the time of the maximum plant height and the initial elevation values at the time of the bare land.

[0010] Preferably, in step S4, the correlation degree screening model expression is as follows: ; Wherein, is the crop growth index leaf area index, is the th vegetation index, is the standard deviation of the crop growth index leaf area index, is the standard deviation of the th vegetation index, is the covariance of the variables and , and is the maximum value; ; Wherein, is the leaf area index fitted by monadic linear regression of the spectral data combined with the ground measured value, and are the first leaf area fitting coefficient and the second leaf area fitting coefficient respectively, is the vegetation index most relevant to the crop growth index leaf area index.

[0011] Preferably, in step S5, the yield fitting model is as follows: ; wherein, is the yield coupling result, the coupling process is to use the unary linear regression to combine the ground measured value of the product of leaf area index and plant height for fitting, the coupling reason is that the leaf area index can represent the horizontal yield of crops, the plant height represents the vertical yield, and the product simulates the overall yield of crops, is the weight coefficient of the product of leaf area index and plant height, is a correction coefficient.

[0012] Preferably, in step S6, the maximum likelihood method, neural network method, support vector machine or numerical classification method in supervised classification is used to extract the salt spot range.

[0013] Preferably, the numerical classification method determines the plot in the set range of the yield coupling result as a salt spot, and the yield coupling result higher than the set range is normal farmland.

[0014] A system for executing the above-mentioned one kind of salt and alkali farmland salt spot extraction method based on multi-spectral unmanned aerial vehicle, comprising: An unmanned aerial vehicle carrying a multi-spectral camera and a positioning module, used for acquiring different spectral images of farmland at different periods; A data processing terminal, used for acquiring image data collected by the unmanned aerial vehicle, and performing image stitching, band calculation and salt spot range extraction.

[0015] Therefore, the present application adopts the above-mentioned one kind of salt and alkali farmland salt spot extraction method based on multi-spectral unmanned aerial vehicle and system, which has the beneficial effects of: By screening the leaf area index most relevant vegetation index, the leaf area index fitted by spectral data is obtained, and the yield coupling result is obtained by combining the plant height data, the salt spot range is extracted according to the yield coupling result, the machine learning algorithm or numerical classification method is used, the algorithm structure model is simple, the yield coupling result is used to identify the salt spot range, which greatly improves the extraction accuracy, and there is no need for complex image processing and coupling, and the practicability is strong, which provides accurate data support for subsequent improvement, fertilization and evaluation.

[0016] The technical solutions of the present application will be further described in detail below through the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flow chart of the present application one kind of salt and alkali farmland salt spot extraction method based on multi-spectral unmanned aerial vehicle; Figure 2Figures (a), (b), (c), (d), (e) and (f) are images of different wave bands of the saline-alkali farmland, wherein (a) is a digital surface model; (b) is a red light wave band image; (c) is a green light wave band image; (d) is a near-infrared wave band image; (e) is a red edge wave band image; and (f) is a visible light image; Figure 3 Figure is a correlation result graph of leaf area index and vegetation index; Figure 4 Figure is a fitting result of leaf area index and vegetation index NDRE; Figure 5 Figure is a principle diagram of extracting wheat plant height in different periods; Figure 6 Figure is a fitting result of plant height; Figure 7 Figure is a fitting result of yield; Figure 8 Figure is a yield distribution graph of the saline-alkali farmland after coupling in the embodiment, and the unit is kg / mu; Figure 9 Figure is a result graph of extracting salt patches, wherein (a) is an actual salt patch distribution graph; (b) is a classification result based on ΔDSM; (c) is a classification result based on NDRE; (d) is a classification result based on visible light; (e) is a classification result based on the coupled yield model (based on samples); and (f) is a classification result based on the coupled yield model (based on numerical classification); Figure 10 Figure is an actual salt patch distribution graph of the whole high-definition image of the saline-alkali farmland in the embodiment; Figure 11 Figure is a salt patch distribution graph of the whole saline-alkali farmland in the embodiment, which is recognized based on the numerical classification method. DETAILED DESCRIPTION

[0018] In the description of the present application, it should be noted that the terms "upper", "lower", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, or the orientation or positional relationship commonly used when the product of the present application is used, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In the description of the present application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "arrangement", "installation", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be connected inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0019] The embodiments of the present application will be described in detail below with reference to the drawings.

[0020] The embodiment is located in a severe saline-alkali land in Lizizha Village, Huanghua City. Four plots are selected as test objects, and the cultivation mode is the conventional winter wheat-summer corn planting mode in Cangzhou. The wheat variety is Hanjianmai 19. DJI Mavic3M unmanned aerial vehicle is used for data collection, which is equipped with multispectral camera (including red light R, green light G, near-infrared NIR, red edge RE, and visible light band spectrum) and RTK device. The shooting is carried out from April 2024 to May 2025.

[0021] As shown in Figure 1 , a salt spot extraction method for saline-alkali cultivated land based on multispectral unmanned aerial vehicle, the specific steps are as follows: Step S1: The unmanned aerial vehicle equipped with a multispectral camera and a real-time dynamic positioning module is used to collect image data of crops in saline-alkali cultivated land at multiple different stages.

[0022] The multispectral camera acquires at least red light band image, green light band image, near-infrared band image and red edge band image; and the real-time dynamic positioning module is provided on the unmanned aerial vehicle, which uses RTK (Real-Time Kinematic) device to acquire the surface model of the crop or cultivated land surface, i.e. elevation information. In this embodiment, the unmanned aerial vehicle uses low flight height and high resolution aerial survey to shoot high-definition image. In order to obtain accurate data, the flight height of the embodiment is selected as 12 m, and the accuracy is 0.55 . In actual application, if the area of saline-alkali farmland is large, the flight height can be appropriately increased to improve efficiency. The maximum flight height should not exceed 130 m.

[0023] The unmanned aerial vehicle is used to shoot at important stages of crops to obtain image data of crops at multiple different stages.

[0024] In this embodiment, the unmanned aerial vehicle data is collected in the saline-alkali cultivated land after the sowing of Hanjianmai and before the emergence, to obtain the initial digital surface model (i.e. elevation data) of bare land, which is convenient for subsequent simulation calculation of the height of Hanjianmai plants. Since the height of Hanjianmai is relatively low, the unmanned aerial vehicle uses tilt flight in three-dimensional flight mode when acquiring elevation data, which can improve the accuracy of collected data. In addition, multispectral and elevation data of Hanjianmai at seedling stage (April 1), filling stage (May 20) and mature stage (May 30) are obtained in sequence. The difference between the digital surface model of cultivated land at different stages and the initial digital surface model of bare land (i.e. the difference of elevation data) is the plant height measured by the unmanned aerial vehicle.

[0025] Step S2: as shown in Figure 2The images collected in each period are spliced to obtain complete saline-alkali farmland spectral images of each band, visible light images and full-range elevation value digital surface models. The UAV image splicing is the process of splicing multiple UAV images into one large image. This process includes image calibration, feature extraction, feature matching, image splicing and fusion steps. The splicing process is a conventional technology and can be automatically spliced by one key through UAV aerial image splicing software (such as DJI ZhiTu, etc.).

[0026] Step S3: Perform band calculation on the complete saline-alkali farmland image data to obtain vegetation index and elevation data in different periods; the vegetation index includes normalized vegetation index, green normalized vegetation index, soil-adjusted vegetation index, optimized soil-adjusted vegetation index, normalized difference red edge index, improved chlorophyll absorption index and green-red vegetation index. The normalized vegetation index calculation formula is as follows: ; Among them, is the normalized vegetation index, is the near-infrared band reflectivity, is the red band reflectivity; The green normalized vegetation index calculation formula is as follows: ; Among them, is the green normalized vegetation index, is the green band reflectivity; The soil-adjusted vegetation index calculation formula is as follows: ; Among them, is the soil-adjusted vegetation index, and are respectively the first fixed coefficient and the second fixed coefficient, which are respectively 1.5 and 0.5, used to suppress the influence of soil background, and the soil is usually 1.5 and 0.5, according to the soil type (ordinary soil, black land or red land, etc.); The optimized soil-adjusted vegetation index calculation formula is as follows: ; Among them, is the optimized soil-adjusted vegetation index, is the standard value representing the canopy background adjustment factor, which is 0.16, and the soil background suppression is optimized; The normalized difference red edge index calculation formula is as follows: ; Among them, is the normalized difference red edge index, For red edge band reflectance; The improved chlorophyll absorption index is calculated according to the following formula: ; Wherein, The improved chlorophyll absorption index, The improved parameter is 0.2 (determined according to soil type, 0.2 for ordinary soil), and the chlorophyll absorption depth is sensitive and resistant to soil interference. The green-red vegetation index is calculated according to the following formula: ; Wherein, The green-red vegetation index.

[0027] As shown in Figure 5 , the plant height data is obtained according to the difference between the digital surface model of the cultivated land in the mature period and the initial digital surface model of the cultivated land in the bare land period. The elevation difference (DSM difference) of all plots in the UAV shooting range is calculated by using the band calculation method. The measured plant height value is fitted by using the DSM difference, and the result is shown in Figure 6 , the upper R 2 =0.7058 indicates that the model can explain 70.58% of the change amount of the plant height. Although the dry alkali wheat height is low, the fitting result of large-scale UAV aerial survey is often larger, and since the tilt flight collection in the UAV three-dimensional shooting is applied in this data collection, the final result has high accuracy, and R 2 The result can reach more than 0.7. The plant height data fitted by the elevation data DSM is fitted by using the linear regression method, and the fitting formula is as follows: ; Wherein, △DSM is the difference between the elevation data DSM of two different periods.

[0028] Step S4: Establish a correlation degree screening model, determine the vegetation index most related to the crop growth index leaf area index according to the correlation degree through the correlation degree screening model, and calculate the crop growth index leaf area index through the most related vegetation index in the previous growth period of the crop mature period (in this embodiment, the grain filling period, for soybean, the pod and grain filling period, corn, the grain filling period, and cotton, the flower and boll period, the judgment standard is that the plant height is maximum and the leaf is not yellow, and the specific day in the grain filling period has little effect on the result accuracy). The expression of the correlation degree screening model is as follows: ; Wherein, The crop growth index leaf area index is the ratio of the total area of all leaf surfaces of the plant to the land area occupied by the plant, which can reflect the potential leaf area of the plant for light interception and gas exchange, and is commonly used to characterize the plant growth. the first vegetation index, the standard deviation of the crop growth index leaf area index, the standard deviation of the first vegetation index, the first vegetation index, the standard deviation of the first vegetation index, the covariance of the variables and the variable is maximized, the Pearson correlation coefficient has a value ranging from -1 to +1, and the closer the absolute value is to 1, the stronger the correlation.

[0029] The ground positioning measurement points in the measurement range were measured, and the leaf area index LAI data was measured during the grouting period. The leaf area index LAI data was measured by an ACCUPAR LP-80 plant canopy analyzer, which is a professional instrument commonly used in agriculture. The data detection device of the instrument is a light energy sensor. The instrument calculates the leaf area index LAI of the canopy by measuring the PAR ratio of the canopy top and bottom. The yield data and plant height H data were measured during the mature period; through statistical correlation analysis and sorting, since the vegetation index is calculated from the spectral bands R, G, B, NIR, and RE, if all participate in model calculation, there will be a collinearity problem, which will lead to unreliable model accuracy, so finally only one vegetation index with the best correlation with the measured LAI is selected for modeling.

[0030] As shown in Figure 3 , according to the correlation results of each vegetation index and the crop growth index leaf area index, the embodiment selects the vegetation index NDRE with the best correlation with the leaf area index LAI, and the correlation coefficient can reach r=0.933, and P<0.01.

[0031] As shown in Figure 4 , R 2 , RMSE, and nRMSE are used as accuracy verification indexes. R 2 is used to measure the goodness of fit between the estimated value and the measured value, and the closer the value is to 1, the stronger the model's explanatory power; RMSE and nRMSE reflect the deviation between the estimated value and the measured value, and the lower the value, the better the model's robustness and consistency. Generally, nRMSE≤10% indicates that the model has excellent consistency, 10%-20% is better, 20%-30% is general, and more than 30% indicates poor consistency. In the above figure, R 2 ​=0.9051, which indicates that the model can explain 90.51% of the variation in leaf area index LAI. In the measurement range, select the position with larger growth gradient change to arrange the measured points, and pay attention to avoid the salt spot range, insert the card or mark other positions, measure the leaf area index LAI data in the filling period, and measure the yield and actual plant height data in the mature period. The fitting formula of leaf area index LAI and NDRE is as follows by using the linear regression method: .

[0032] Step S5: Establish a yield fitting model, couple the crop growth index leaf area index and the corresponding elevation data through the yield fitting model, obtain the yield fitting data and convert it into a visual result. During the growth of crops on heavy saline land, there is basically no crop growth in the salt spot range, even if there are individual weak seedlings struggling to grow, they cannot mature and form yield. Therefore, yield of 0 or very small can be equivalent to being in the salt spot range, so as to quantify the salt spot index and facilitate extraction.

[0033] In step S5, the yield fitting model expression is as follows: ; Among them, is the yield coupling result, and the coupling process is to use linear regression and the product of leaf area index and plant height to combine the ground measured value for fitting, and the coupling reason is that leaf area index can represent the horizontal yield of crops, and plant height represents the vertical yield, and their product represents the overall yield of crops, is the weight coefficient of the product of leaf area index and plant height, is a correction coefficient. As shown in Figure 7 , according to the fitting result, 214.19 is taken, 52.694 is taken.

[0034] The fitting model result is widely promoted and visualized by band calculation, as shown in Figure 8 .

[0035] Step S6: According to the visual result as the bottom map data of image classification, automatically extract the salt spot range according to Figure 8 combined with machine learning algorithm.

[0036] The specific operation is to use image analysis software such as ENVI image processing software, select supervised classification under image classification, and select one of maximum likelihood method, neural network method, minimum distance method, support vector machine and the like in the classification algorithm.

[0037] The maximum likelihood method, neural network method or support vector machine algorithm in supervised classification can be used for salt spot extraction. In order to compare the classification effects of different base maps, the machine learning method based on training samples is selected in this embodiment, and the support vector machine algorithm is used for salt spot range extraction. In addition, since the final coupling result is yield data, the yield of 0 or a very small amount can be equivalent to being within the salt spot range, and the salt spot index can be quantified. The salt spot range is obtained by quantifying the salt spot index. This method is more efficient, more accurate and does not depend on sample training. The salt spot extraction value set during classification in this embodiment is: the land with a coupling yield value in the range of 0-50 kg / mu is extracted as a salt spot.

[0038] In order to verify the superiority of this embodiment, other existing methods are compared. Three control groups are classified by visible light, ΔDSM and NDRE as the classification base map. At the same time, since the coupling yield has actual numerical significance, the numerical classification method is used in this embodiment, and the comparison results of each method are shown in Table 1 and Figure 9 .

[0039] Table 1 comparison results

[0040] The most common method is to directly apply various classification models of different algorithms to image classification based on visible light base map. However, since the visible light band has more wave bands (at least three wave band information), the data contained is complex, so it is more dependent on complex algorithms to improve accuracy, and the time used is very long. As shown in Figure 9 , the pink translucent area is the separated salt spot range, compared with the actual salt spot distribution map (a) shown in Figure 9 , the base map classified by ΔDSM (b) shown in Figure 9 is easy to divide the farmland around the salt spot into salt spot, and the base map NDRE classification (c) shown in Figure 9 is more likely to miss the salt spot. The classification of base map visible light (d) shown in Figure 9 and coupling yield (e) shown in Figure 9 is acceptable in the figure, but visible light has more redundant information, and many small spots can be seen obviously, which can be referred to as visible light classification (f) shown in Figure 9the upper right corner of Fig. 6 (d), which is the main reason for the low accuracy and precision of the visible light classification. Even based on the longest time, the classification accuracy is only 0.8041. Since the method of the present embodiment filters out the most critical spectral band information, eliminates spectral redundant bands, and couples the yield results closely related to growth, the final classification base map information content is small, but closely related to salt spot extraction and easy to calculate (only contains yield information of one band). Therefore, when using the same classification method, the support vector machine algorithm based on training samples, the accuracy and precision of the classification are higher than those of other classification base maps, the edge is smoother, the graph is more complete, and there are fewer small graph patches, as shown in Fig. 6 (e). Figure 9 The time used is similar to that of other single-band base maps ΔDSM and NDRE, about 1 hour, and accurate samples still need to be provided manually to ensure the accuracy.

[0041] Based on the coupled yield base map, the land with yield value in the range of 0-50 kg / mu is extracted as a salt spot, and a numerical object-oriented classification method is used for salt spot extraction. The results are shown in Fig. 6 (f) and Table 1. Figure 10 The classification accuracy and precision are improved, and the classification time is greatly shortened, only 15 min, much lower than other methods, and no samples need to be provided manually, only two yield value limits of 0 and 50 need to be input, and the salt spot graph patch with high accuracy can be automatically extracted, which greatly improves the workload and classification efficiency.

[0042] Based on the above classification method, compared with high-definition visible light images (Fig. 7), Figure 11 Through the extracted salt spot data, the salt spot position distribution map in the selected test range of the present embodiment can be obtained, as shown in Fig. 8. ​The salt patch range distribution obtained by the method can be used for saving the amount of fertilizer and pesticide when fertilizing and spraying pesticide in saline-alkali farmland. For example, in the range of the embodiment, the specific operation is as follows: during the crop cultivation process, when unmanned aerial vehicle is used for fertilizing or spraying, the salt patch range is erased in the operation path, that is, fertilizing and spraying are not needed in the salt patch range, and the unmanned aerial vehicle sprays according to the erased range, so that the input of chemical fertilizer and pesticide is saved, and the environmental pollution is reduced. In the embodiment, the total area of the plot is 104.43 mu, and the area occupied by the salt patch is 19.76 mu. After removing the salt patch range, 18.9% of the pesticide and fertilizer can be saved, and the real fine management of the land is realized.

[0043] Similarly, according to the salt patch range distribution obtained by the method, a spraying unmanned aerial vehicle can be used to spray biological agents, medicines, special organic fertilizers for saline-alkali land and other saline-alkali land improvement agents only in the salt patch range, so as to realize accurate coverage of the salt patch and treatment of saline-alkali land according to the disease. For example, in the embodiment, the improvement agent cost can be saved by 81.1%, better improvement effect can be achieved at a lower cost, or the concentration of the agent can be further improved without worrying about destroying the microbial balance in the normal soil.

[0044] In terms of accurate yield estimation, accurate statistics of the salt patch range and area can accurately calculate the effective planting area of the actual yield, so as to obtain more accurate yield per mu, which is more meaningful for evaluating the harvest and the improvement effect of saline-alkali farmland.

[0045] A system for performing a multi-spectral unmanned aerial vehicle based saline-alkali farmland salt patch extraction method, comprising an unmanned aerial vehicle and a data processing terminal.

[0046] The unmanned aerial vehicle is provided with a multi-spectral camera and a positioning module, and is used for acquiring different spectral images of farmland at different periods.

[0047] The data processing terminal is used for acquiring image data collected by the unmanned aerial vehicle, and performing image stitching, wave band calculation and salt patch range extraction.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle (UAV), characterized in that, The specific steps are as follows: Step S1: Use a drone equipped with a multispectral camera and a real-time dynamic positioning module to collect image data of crops at different stages in saline-alkali farmland. Step S2: Stitch together the images collected in each period to obtain a complete image of saline-alkali farmland; Step S3: Perform band calculations on the complete saline-alkali farmland image data to obtain vegetation index and elevation data at different times; Step S4: Establish a correlation screening model. Based on the correlation degree, determine the vegetation index most correlated with the crop growth index leaf area index using the correlation screening model. Calculate the crop growth index leaf area index by fitting the vegetation index most correlated with the previous growth stage before the crop maturity stage. Step S5: Establish a yield fitting model, and couple the crop growth index leaf area index and the corresponding elevation data through the yield fitting model to obtain yield fitting data and transform it into visualization results; Step S6: Based on the visualization results as the base map data for image classification, the salt spot range is automatically extracted using machine learning algorithms.

2. The method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle according to claim 1, characterized in that: In step S1, the multispectral camera acquires at least red band images, green band images, near-infrared band images, and red edge band images; and the UAV is equipped with a real-time dynamic positioning module to acquire the elevation values ​​of the crop or cultivated land surface. Drones were used to photograph saline-alkali farmland before seedling emergence and during the growth period before crop maturity to obtain the initial digital surface model of bare land, the initial elevation value when the land was bare, the elevation value when the plant height was at its maximum, and the vegetation index during the growth period before crop maturity. To improve the accuracy of elevation values ​​when acquiring digital surface models of saline-alkali farmland, drones were used to collect elevation values ​​at the maximum plant height and when the land was bare, employing an inclined flight method.

3. The method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle according to claim 2, characterized in that: In step S3, the vegetation indices include the normalized vegetation index, the green normalized vegetation index, the soil-adjusted vegetation index, the optimized soil-adjusted vegetation index, the normalized differential red edge index, the improved chlorophyll absorption index, and the green-red vegetation index.

4. The method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle according to claim 3, characterized in that: Plant height data were calculated based on the difference between the elevation value at which the plant height reached its maximum and the initial elevation value when the land was bare. The data was then fitted using measured plant heights from fixed-point samples. The fitting calculation formula is as follows: ； in, The fitted plant height data, and These are the high-fit coefficients for the first and second plants, respectively. This is the difference between the elevation value at which the plant height reaches its maximum and the initial elevation value when the land is bare.

5. The method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle according to claim 4, characterized in that: In step S4, the relevance screening model expression is as follows: ； in, Leaf area index is a crop growth indicator. For the first Individual vegetation indices, The standard deviation of the leaf area index, a crop growth indicator. For the first The standard deviation of each vegetation index For variables and variables covariance, To find the maximum value; ； in, The leaf area index is obtained by fitting spectral data. and These are the fitting coefficients for the area of ​​the first leaf and the area of ​​the second leaf, respectively. The vegetation index is most correlated with the leaf area index, a crop growth indicator.

6. The method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle according to claim 5, characterized in that: In step S5, the yield fitting model is as follows: ； in, As a result of production coupling, This is the weighting coefficient for the product of leaf area index and plant height. This is a correction factor.

7. The method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle according to claim 6, characterized in that: In step S6, the salt spot range is extracted using methods such as maximum likelihood estimation, neural networks, support vector machines, or numerical classification in supervised classification.

8. The method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle according to claim 7, characterized in that: The numerical classification method identifies plots with yield coupling results within a set range as salt patches, and plots with yield coupling results above the set range as normal arable land.

9. A system for performing the method for extracting salt spots from saline-alkali farmland based on a multispectral unmanned aerial vehicle as described in claim 8, characterized in that, include: A drone equipped with a multispectral camera and a positioning module is used to acquire spectral images of farmland at different times. The data processing terminal is used to acquire image data collected by the UAV and perform image stitching, band calculation, and salt spot range extraction.

Images